CMEIAS Color Recognition Operator Manual
Contents
1. Fig 17 Applying the Rotate Image Clockwise feature to an image Rotate Clockwise Angle degrees 25 k Save Sampled Pixels This time saving feature allows the set of digital information for the sampled pixels to be saved to a txt file for future use to facilitate the color segmentation of multiple images whose foreground objects have a similar or the same color range The saved file has a tab delineated format that allows ease of pasting into spreadsheets if so desired When this option is chosen a dialog box displays asking you to enter the text filename and location where the digital information for the set of sampled pixels will be saved Only one set of sampled pixels exist while the program is running When a user samples a pixel its color and location are added to the set of sample pixels that can be resaved Example of digital information saved in the text file of 3 pixels sampled from an image 3Sample_Pixels txt user specified file name X Y R G B 314 261 255 3 3 43 310 255 8 5 196 245 255 3 2 l Load Sampled Pixels This time saving feature allows you to apply the color segmentation routine on an active image using the saved text file see section IV B 4 k above containing the digital information on the set of pixels previously sampled from foreground objects with similar or the same RGB range in a different image Once this option is chosen a dialog box displays to spec2. Color Similarity Tolerance 65 gt OK Discard any sampled pixels in memory al Select 2 pixels from 2 red cells within the input image Apply the color segmentation function Process Apply Color Segmentation amp If necessary apply 1 cycle of the erosion dilation process Process Apply Color Dilation Ea Process Apply Color Erosion Ei Y NY a 10 11 12 36 Before gt After image L f g G 4 W a a u 7 ES lr oe Tdi _ y Segment the Gram negative spiral bacteria in Image2 tif Open Image2 tif Set the background color to white Process Segmented Image Background White Apply contrast Filter Apply Contrast Set Color Similarity Tolerance to 100 pixels Process Color Similarity Tolerance enter 100 OK Discard any sampled pixels in memory il Load previously saved set of RGB values for the sampled foreground pixels of the red cells Process Load Sampled Pixels Select Image2pixels txt Open Apply the color segmentation function Process Apply Color Segmentation b Note that background pixels within the same color range are included in the segmentation output image prepared using the above settings These are derived from the red colored pixels of the halo surrounding the Gram positive bacteria in the same image Optimize the MinMax Size Filter settings to remove the residual background pi
3. d Paste Shortcut Ctrl V The Paste command displays the TIF BMP or PNG image from the Clipboard into a new image window One can save this pasted image under a new name and or location 3 VIEW a ToolBar The Tool Bar feature displays the Toolbar below the main menu Fig 5 and contains shortcut icons for many common operations When the mouse is positioned over an icon in the toolbar the status bar at the bottom of the main window will immediately indicate what function that icon provides When the cursor is held for a second over any icon a tool tip indicating the action it performs will appear below the mouse cursor b Status Bar The Status Bar at the bottom of the main graphical user interface window Fig 5 displays various information about the active image including its zoom ratio the x y coordinates 0 O landmark origin in the upper left corner and the RGB brightness intensities of the pixel located precisely at the cursor position the number of pixels sampled to operate the color segmentation algorithm the Time s with millisecond precision required to load the input image its width x height pixel dimensions and the number of bits required to store a single pixel 8 bits for grayscale 24 bits for RGB The brightness intensities will be independent values between 0 255 for each RGB chromatic channel for color images and be the same RGB values 0 255 for grayscale images 16 When a previously saved file
4. 22 ABC A third use is to help define the edges of cells of interest when a fluorescent halo extends from outside their contour into background Fig 23 The output image of this tool can be saved and used to reduce and or eliminate the fluorescent halo around fluorescent cells in order to produce the final image for quantitative image analysis Fig 21 Use of the Find Object Edges filter to quickly find areas of the image where foreground objects are touching each other Examples are indicated by white arrows q Y Copy 7 of Imag El X F Copy of Copy Jo Figure 22abc Use of the Find Object Edges filter to isolate foreground objects for copying into another image A original multicolored image B color segmented output image of green cells C Output image of Find Object Edges filter applied to the green color segmented cells with separation of foreground and background pixels that can be further processed Figure 23 Use of the Find Object Edges filter to define the contour of cells with a fluorescent halo k Emboss This image processing feature produces a three dimensional textured perspective of foreground objects in the active image It does so by brightening one side and darkening the opposite side Fig 24 This filter can reveal variations in pixel brightness of internal structures and provide perspective on object thickness and surface texture Fig 24 The Emboss filter is use
5. MMOS a o E ala daa SS 14 Ds RECO as de du a A E A Ses eae a 14 Es COPY dale wae cab fae d ih e A A fe Ge ir o ona a E 14 O Rasta Si E aaa RUE E A 14 Bs VIEW Sd da e A a E E O E 14 de COOL BAR s gf 15 ma TE o GDE sa MA VE A BO E 2 E 14 Ds Status BAR s v dp a a SD a E 14 CG Fitto SerGe n a sd E a e E a e a a GE a 15 Gs ZOOM DAS 2d 4 4 0 book DU A E EU A D 15 e Normal Viewing QUA a sa i aaa e eS aS ea E 15 Zoom UE ss msg DE Ed AMRS DE A eR E AS 15 go Select Pixel Sampling Cursor w r ps mi RS REDE E E we eee 15 We OC OSS cor qu quo he tat iat ly e RR e Re WE Dee im E DD A o AAA O E 16 a Segmented Image Background noa moa a 16 b Color Similarity Tolerance s 1 ke ee a 16 C Apply Color S ementation sm 4 k dd oe Se a ee E 18 a Apply Color Dilation sis aee se grs E RS DC SP RL Se ES E 18 e Apply Color Erosion asa sas pos GA sa ea ee 19 MUSMALAOC Lami CNE NT Be dr ee Sw 20 6 KolorModelS de 21008 Ed E Bela GD Eco oe RAE Sd 20 iy Spntito RGB Channels sm de de eat ke a ak Gh eS 20 Ze OPULO YUV Channels e a era wie ia REA im E 6 21 3s ODUE TOM SL Channels s ppp DES ES eh SG 22 4 Convert to PSeudocolorss s amp ai A do aa So E 22 h FUp Mates aem Gea A SO a ae A ee a ae Gn ee 23 lo HOFIZOMtAL ie fase we ob tek AA E ot A a 23 Zs VETUCAL a a wou sa Bate A A A 23 L Negativo Maga lt i ais we e i Gs ee A a A A 23 j Rotate Image Clockwi
6. a different size This menu item is grayed out when the image is first loaded since it s already ata 1 1 size 100 zoom f Zoom Out Shortcut Ctrl The Zoom Out command adjusts the zoom of the active image window to a lower displayed magnification ratios of 1 2 1 4 1 8 and 1 16 g Select Pixel Sampling Cursor 80 vel This feature provides six different cursor choices to optimize the accuracy of the interactive pixel sampling process for a wide range of images The default choice is the pointed tip of the Window s standard arrow far right example in the above image 17 4 PROCESS a Segmented Image Background This option allows the user to select a white default or black colored background for the color segmentation output image Make that decision based on the type of microscopy used to acquire the image and type of object analysis to perform on the fully segmented image e g black for epifluorescence white for transmitted brightfield or phase contrast b Color Similarity Tolerance Color Similarity Tolerance XxX Li r Figure 6 Dialog box displayed to specify the threshold setting that defines the Color Distance Units _ 1 400 100 color similarity tolerance for the color segmentation function Process Color Similarity Tolerance The unit entered represents the radius of all spheres in 3d color space of included colors whose center 12 the sampled pixel This process adjusts t
7. are obscured Fig 15 24 h Flip Image 1 Horizontal This tool flips the active image horizontally along its vertical axis 2 Vertical This tool turns the active image upside down by flipping vertically along its horizontal axis i Negative Image The Negative Image command converts the brightness value of each pixel s corresponding channel s in the original positive image to its inverse value in the 256 step calibration to produce the negative output image Fig 16 For an RGB color image example a pixel in the positive image with RGB values of 35 203 48 is changed to 220 52 207 For a grayscale image example a pixel in the positive image with a brightness value of 255 is changed to 0 and a pixel with a brightness value of 25 is changed to 230 This tool works on both RGB color and grayscale images Fig 16 Fig 16 Inversion of the positive color and grayscale images using the Negative Image tool j Rotate Image Clockwise This option applies a clockwise rotation of the active image by a user defined angle When selected a dialog box displays to input any rotation value in degrees between 1 and 360 default is 90 Fig 17 To increase the precision the rotation angle can be specified with float values with decimal points This option is useful to make the image exactly horizontal when acquired from a twain device RF ImagelA tif
8. bit RGB it is first converted to the corresponding grayscale image using the algorithm described in section 5a above and then the grayscale image is automatically converted to the binary image by the threshold operation using the same specified brightness level The Training Tutorial Appendix 1 illustrates the use of the brightness threshold together with the Split to YUV model to segment colored objects 27 c Adjust Hue Saturation The hue saturation and intensity characteristics of the RGB color wheel are described in Section IV B 4 g 3 entitled Split to HSI and its accompanying Fig 14 The Adjust Hue Saturation filter allows the user to adjust the hue saturation and lightness of the entire active RGB image prior to selecting the foreground pixels of interest for color segmentation When selected it opens a dialog box with two inputs one to enter the hue value and the other to enter the saturation value Fig 19 The default values are arbitrarily set at 12 and 50 respectively An adjustment of the saturation level purity of the color represents a move across the radius of the color wheel vector A whereas an adjustment of the hue color represents a move around its perimeter vector B Fig 19 Figure 19 Dialog box for the Adjust Hue Saturation filter and the Adjust Hue Saturation Hue rE color wheel showing the saturation A and hue B vectors Saturation 50 Cancel d Use the Adjust Hue S
9. emeiasfd msu edu GE e Enjoy CMEIAS 41 Additional training not included in the CMEIAS Color Segmentation AV tutorial mM E ea E Segment the Green fluorescent cells in Image1 tif Open Image1 tif 5 Set the background color for the output image to black Process Segmented Image Background Black While viewing the RGB values for pixels of green cells note that they include intensities in both red and green chromatic channels but no values of blue Set the Color Similarity Tolerance setting to 65 pixels Process Color Similarity Tolerance 65 OK Discard any sampled pixels in memory al Select 2 pixels from 2 green cells within the input image Apply the color segmentation function Process Apply Color Segmentation de If necessary apply 1 cycle of the dilation erosion process Process Apply Color Dilation E Process Apply Color Erosion z II 42 Before gt After image Segment the Gram positive rods in Image5 tif Open Image5 tif This image poses several challenges for successful segmentation Because of the very noisy background a sufficiently large number of training pixels must be sampled from various Gram positive bacteria within the image 153 sampled pixels of foreground objects are included in the Image4pixels txt file Zoom out to 1 2 ratio Ey Set the background color to white Process Segmented Image Background White Set the color
10. enhancing the utility of the software application Min Max Object Size Filter gt Generate Preview Area Apply Size Filter to Image Smoothen Sharpen Object Edges Find Object Edges Emboss Preprocessing features include options to work with a duplicate image while comparing processed steps to the original split the image to its RGB YUV or HSI color channels specify the color similarity tolerance threshold levels and output image background color adjust image intensity contrast and find smoothen sharpen object edges Post color segmentation functionalities include a user defined minimum maximum object size filter to discriminate the size range of foreground objects included in the output image a feature to fill small holes of lost pixels completely enclosed within foreground objects and a color dilate erode feature to eliminate residual background noise in the output image and compensate for object segmentation with imperfect parameter settings These features handle common remaining segmentation problems inherent in fluorescence micrographs e g where the curvature of strongly fluorescent cells creates halos of similarly colored pixels and or fluorescent cells have dark internal areas at lower sampling density A useful option to save and retrieve the color range of selected training pixels is also featured 13 to add semi automation capability to the color segmentation routine when many image samples of the same community a
11. holes in grayscale images can also be filled using this feature after the color segmentation routine has been applied oleae Vee Fle ee Vleet Fill Small Holes Fig 11 Sequential use of the Fill Small Holes post processing feature to convert internal Maximum size of holes filled it pixels foreground pixels lost during color nam segmentation into the object s 30 Hole Size E average color Arrows point to the position where holes were filled by applying the filter at the indicated maximum hole size pixels setting g Color Models 1 Split to RGB Channels This feature will split the active image into three new separate windows indicating the Red Blue and Green chromatic channels Fig 12 In some cases changing a color image into a single channel can facilitate its segmentation of the objects of interest but more commonly the objects contain colored pixels with two or three channels Fig 12 This feature helps to diagnose complex segmentation problems in color images The recommended solution is to use the Apply Color Segmentation routine featured in this software Applying this feature to a grayscale image will display its 3 RGB channels of equal brightness 22 Original Image W Red Channel R Green Channel Blue Channel Fig 12 Splitting of an RGB image into its Red Green and Blue chromatic channels 2 Splitto YUV Channels This feature splits the active RGB image bas
12. of sampled pixels is loaded see sections IV B 4 k and IV B 4 1 the number of pixels previously sampled in that saved file will display If you select more pixels after loading the file of previously saved pixels then this number will increment to include the newly sampled pixels The status bar will also display the time millisecond precision required to run the color segmentation algorithm and display its resultant output image When the cursor is positioned over a toolbar icon or menu item the Status Bar displays context sensitive information about its function instead of the zoom x y coordinates pixel brightness values c Fit to Screen The Fit to Screen command automatically changes the zoom level so that the active image can fully display within the graphical workplace on the monitor screen while maintaining its aspect ratio Before using this option you must maximize the image window When this option is selected the three other zoom options described below are inactive d ZoomIn Shortcut Ctrl The Zoom In command adjusts the zoom of the active image to a higher displayed magnification ratios of 2 1 4 1 8 1 and 16 1 You can select to view the magnified area of the image by using the horizontal and vertical scroll bars e Normal Viewing 1 1 14 Shortcut Ctrl The Normal Viewing command returns the display of the active image to its 1 1 original size of view after previously being zoomed to
13. places this file within the same folder as the executable CmeiasColorSegmentation exe program file c Help Topics This command displays the contents of the CMEIAS Color Segmentation help system From its table of contents you can select the topic providing the information you need Keywords can be entered from the Search tab to produce a list which when selected will display the corresponding page with the keyword highlighted within the text 32 d CMEIAS Website This command opens the home page of the CMEIAS Project website in your computer s default browser The website url address is http cme msu edu cmeias The Color Segmentation webpage can be accessed by clicking its Hot button displayed among other CMEIAS features on each webpage Check it and the CMEIAS News page periodically for pertinent information updates and new version releases Download amp Register amp Citation 1 D Object Color Ee Introduction installation Contact Us Info Classification Segmentation User Object 14 D Morph Publications Support Contributors analysis Classification Using CMEIAS CMEIAS Center for Microbial Ecology Image Analysis System Requirements CMEIAS Project Mission Statement Develop amp release a comprehensive suite of bioimage informatics analysis software applications designed to strengthen quantitative microscopy based approaches for understanding microbial ecology at spatial scales relevant to th
14. the status bar that range approximately between 200 230 39 8 Apply the Brightness Threshold filter at a setting level of 190 Filter Brightness Threshold enter 190 OK 9 Invert the brightness levels of this thresholded image using the Process Convert to Negative Image function to produce the final segmented image containing only the original blue cells The Before After images display below Before gt After image 10 Next we will isolate the green cells from the other duplicate U channel image 11 Measure the brightness values of the darkest gray cells originally green by positioning the cursor over them and view their gray levels displayed in the status bar The brightness values for the pixels of those cells range approximately between 33 43 12 Apply the Brightness Threshold filter using a setting level of 60 Filter Brightness Threshold enter 60 OK 13 Remove any background pixels remaining in the output image using the E e erosion dilation process The before and after images are shown below E Before gt After image 14 Finally we will isolate the red cells which are the bright ones in the V channel image To begin duplicate the V channel image 15 16 40 The bright cells originally red can be isolated from this image by applying the Brightness Threshold process using a setting level of 190 Filter Brightness Threshold enter 190 OK Conv
15. 1 CMEIAS Color Segmentation User Manual Colin A Gross Chandan K Reddy amp Frank B Dazzo Center for Microbial Ecology Michigan State University 2010 About Cmeias Color Segmentation a SD xj Color Ses mentat on o o E 4 pete Heme msu FATES _ Copyright c Michigans State unien CMEIAS Center for Microbial Ecology Image Analysis System Table of Contents I CMEIAS Color Segmentation License Agreement 84 5 II Background Information 1 1 we ee ee 6 III General Color Segmentation Protocol 46 5650 28 8 2 wee 8 IV Ment S ERUCLUTE g o e A AA O a E Be a A 11 Ao OYE VIC p it Sh Cd Y AA ASA A A 11 B MentrODUIORS oa mo a 2 SE A A E A AA E A A 12 Mis TUG Dr a e Ada 12 da VOOM iia Bs chs ath an hy Ge dada 12 De DAVE s ae tas A AAA 12 E SAV CVS e oe i AE SO q DRE a o O A SR 12 d Duplicate the Active Image a s i os ee ee es 12 e Close Active Made am de md ee ee eg ae a 13 L Acquire Made 5 isi ERA Gide Se OR a 13 T Select Wain SOU ja as ae Quid Ee Ed A DA E 13 2 Acquire Twain Image aoao ee eee es 13 g Print Preview Active Image 2 2 2 2 ee 13 H PME SE UD ps fe ae ide Aloe Se Ge ES EEE EA 13 E PRIMUACIVE Mage md iris bd te 13 Ja JRECENtMIICS sms DME DES ee E He ER 13 Ro a o e SUN DE RE DESCE E De es oe 13 2a BCIE o ses eh ve E E SS SEE DO Des A AE DE DR 14 ds
16. CMEIAS Color Segmentation program segments RGB color images based on interactive sampling of pixels representing the RGB range for the foreground objects of interest followed by mathematical computations of the color information weighted by a user defined threshold setting of their similarity in spatial distances By including this 10 weighted similarity measurement our system provides the flexibility to adapt to several different color groups for the segmentation process even with complex backgrounds The segmentation algorithm is then applied to produce the output image containing the segmented foreground objects of interest against a noise free pure black or pure white background These functions all translate to a reduction in user s time and labor costs required to perform this essential image editing step hence facilitating the whole process of digital image analysis As with all applications of digital image analysis the original color images of the microorganisms must be of high quality as a prerequisite Open Image Brightness Color Models Hue Pre Process If Needed Saturation Intensity Negative Contrast Sharpen Sooothen y Inspect Color Heterogeneity Zoom 1 2 x 12 y 23 RGB 28 7 0 ili ariy Tolerar y Color Distance Units ns o 1 400 Set Color Range Tolerance Moo The unit entered represents the radius of all spheres in 3d color space of included colors whose center is the sampled pixe
17. ation on images whose background pixels are of similar color but differ in intensity Black arrows pointing to the right show the progression of image processing iterations applied at lower similarity tolerance levels to achieve final segmentation Image from Gross et al 2009 In B 2 cells in an anaerobic bioreactor community are stained by fluorescence in situ hybridization FISH using a 16S rRNA phylogenetic probe for Clostridium sp This image was very noisy containing foreground pixels afloat in a background of similarly colored pixels having a different intensity pattern typical of random static interference Accurate color segmentation was achieved by initially setting a high color tolerance threshold level that included all the foreground pixels and then gradually excluding background pixels from the result image by subsequent iterations of the color segmentation routine at progressively lower tolerance levels In C autofluorescent pigments allow detection of cyanobacteria in an estuary sample The accurate desired result was achieved by selecting two training pixels yellow arrows from the colored regions of foreground objects and then performing automated segmentation iterations at decreasing color tolerance threshold levels starting at 160 followed by 105 85 and finally 75 to gradually delete background pixels of autofluorescent detritus and smoothen each cell s contour Also note how touching cells can be separated green arro
18. aturation filter option to add color to a grayscale image and convert it to RGB or to make an RGB image look like a duotone by reducing its color values to one hue Sometimes this adjustment will facilitate the segmentation of foreground colors of interest in color segmentation d Increase Intensity plus The Increase Intensity command increases the intensity of the image i e increases the luminosity or brightness factor in the HSI model Fig 14 vector C Each subsequent use of this option moves all of the pixels of the image closer to white maximum lightness This can be useful if background of an image is non homogeneous and already close to white e Decrease Intensity minus The Decrease Intensity command decreases the intensity of the image i e the brightness factor of the image is reduced This filter is useful for processing images when all but just a few scattered pixels of the background are nearly black In cases such as this the background can be made completely dark Be cautious when using this function as it may reduce the distinction of contrast between foreground and background objects f Add Contrast The Add Contrast filter makes simple adjustments to the tonal range of every pixel in the active image Sometimes this processing step can improve the subsequent sampling of pixels for color segmentation This command does not work with individual channels and is not recommended for high end output because
19. ble from and shall in no way affect the validity or enforceability of the remaining provisions of this Agreement This Agreement shall be governed by Michigan law II Background information Microscopy and digital image analysis are important investigative tools in microbial ecology that provide direct quantitative information on the microbes world from their own perspective and spatial scale without the need for their laboratory cultivation Unfortunately much less information has been obtained from images of microbes than can actually be extracted using computer assisted microscopy primarily because digital images of microorganisms in their natural habitats are highly complex posing major challenges of image processing required for quantitative image analysis An essential and most difficult task is object segmentation which represents all editing steps required to reduce the image to the foreground microbial objects of interest before analysis Complexity of image segmentation is increased even further when the organisms are colored to reveal important information on their ecological biochemical physiological cytological and or phylogenetic characteristics in situ Fig 1 As a consequence information on the richness abundance metabolic activity and spatial heterogeneity of microbial populations and communities in complex environments is often visually described but rarely quantitated from true bitmap color images compromising the poten
20. crobial populations and communities sources of error and how they are addressed and examples of its application to solve various complex image processing challenges commonly encountered in color images acquired for quantitative microbial ecology studies This algorithm s run time depends on the image size heterogeneity of color pixels number of training pixels sampled and speed of the computer The time with millisecond precision required to complete the segmentation algorithm and produce the color segmented output image is displayed in the status bar Figure 8 Original A and color segmented output images of the red B and green C fluorescent bacteria In this example the segmented image background was set as black d Apply Color Dilation ES This command applies a dilation filter to the active color segmented image resulting in a slight enlargement of the foreground objects background size diminishes This filter Operates by converting all background pixels that have at least 1 foreground pixel neighbor to the average foreground color It can be applied to RGB or grayscale images This and the Color Erosion filter are mainly applied to correct certain types of errors in the color segmented output image made by the Apply Color Segmentation routine For example use the color dilation filter to regain foreground object pixels in the image when they have been erroneously deleted called a false dismissal error during color
21. d to make tangential pseudo shadowcast output images from 24 bit RGB top and 8 bit grayscale bottom images The angle of shadow is reversed in the rightmost inverted images using the Convert to Negative Image filter 31 6 WINDOW a Cascade Cascading resizes and staggers layers of all the open image windows within the workspace below the main window so that each title bar is visible b Tile Tiling resizes and arranges all the open image windows side by side in the workspace below the main window c Close All This command closes all the windows that are opened in the main window d Windows List This feature displays the list of images windows that are present in the main window The check mark denotes the active image Any other window can be made active by just clicking on the corresponding image in the list 7 HELP a About Cmeias Color Segmentation This command displays the About shield that also temporarily displays when launching the program shown on the 1 page of this User Manual This image provides information on the current version and authors of the CMEIAS Color Segmentation software the CMEIAS homepage website url copyright information the CMEIAS logo and the desktop shortcut icon for this software b User Manual This command opens the CmelasColorSegmentation pdf file of this user manual document in your default program e g Adobe Reader assigned to display pdf files The software installation
22. e cells in Image4 tif Open the Image4 tif training image Make two duplicate images and close the original Activate one duplicate image and apply the RGB color model to examine its complexity Process Color Models Split to RGB channels The split images should look like the ones shown above Note that the cells in the blue channel are segmented OK but the cells in the red and green channels have pixels of overlapping color and therefore are not well segmented Activate a second duplicate image and apply the YUV color model Process Color Models Split to YUV channels The output images should look like the ones shown below left to right channels Y U and V Before gt After image The Y channel image output is a grayscale equivalent of the original image The image outputs of the U and V channels represent a transformation that discriminates the red green and blue cells based on each group s brightness level Here we will illustrate how this discriminating luminosity can be used in conjunction with the Adjust Brightness Threshold tool to segment the 3 groups of cells Make 2 duplicate images of the U channel image They will be used to isolate the originally blue bright cells and the originally green dark cells from the U channel image Next position the mouse cursor over the bright cells originally blue in one of the duplicate U channel images and note the grayscale brightness values of its pixels in
23. e individual microbes and their ecological niches 33 8 References Reddy C K Feng I Liu and F B Dazzo 2003 Semi automated segmentation of microbes in color images In Color Imaging VIII Processing Hardcopy and Applications Proc International Society for Electronic Imaging SPIE 2003 R Eschbach and G Marcu eds vol 5008 548 559 DOI 10 1117 12 472024 http dx doi org 10 1117 12 472024 and http spiedl aip org getabs servlet GetabsServlet prog normal amp id PSISDG0050080000 01000548000001 amp idtype cvips amp gifs yes amp ref no Reddy C K and F B Dazzo 2004 Computer assisted segmentation of bacteria in color micrographs Microscopy amp Analysis 18 5 7 September 2004 issue Gross Colin A Chandan K Reddy and Frank B Dazzo 2009 CMEIAS color segmentation an improved computing technology to process color images for quantitative microbial ecology studies at single cell resolution Microbial Ecology online version DOI 10 1007 s00248 009 9616 7 Gross Colin A Chandan K Reddy and Frank B Dazzo 2010 CMEIAS color segmentation an improved computing technology to process color images for quantitative microbial ecology studies at single cell resolution Microbial Ecology 59 2 400 414 Enjoy CMEIAS Frank Dazzo cmeiasfd msu edu 34 Appendix 1 Training Tutorial amp Images CMEIAS Color Segmentation Training Tutorial by y Frank B Dazzo Michigan State University This appendi
24. ed on its YUV content This is a color encoding scheme for natural pictures in which luminance and chrominance are separate The human eye is less sensitive to color variations than to intensity variations YUV allows the encoding of luminance Y information at full bandwidth and chrominance UV information at reduced bandwidth The Y channel image is a color to grayscale conversion The U channel maximizes the contrast between dark green cells and bright blue cells against the gray background and the V channel maximizes the contrast between bright red cells and dark green cells against the gray background Fig 13 Such changes in contrast can be used in conjunction with brightness thresholding procedures to segment the colored cells The Appendix I training tutorial includes the use of this feature to achieve color segmentation E Original Image 3 Y Channel Y U Channel E Y Channel Fig 13 Splitting of an RGB image into its Y U and V channels The Y channel image is a grayscale conversion from the original image The U and V channels have contrast adjustments that increase differences in luminosity between the blue and green cells and between red and green cells respectively 23 3 Split to HSI Channels Fig 14 Vectors of the RGB color wheel A Saturation B Hue C Brightness D All hues Source Adobe Photoshop Help Center Based on the human perception of color the HSI model describes three fundamental characterist
25. ert the image to negative Process Convert to Negative Image followed by one cycle of erosion dilation to produce the final output E Es image with the original red cells now fully segmented in a noise free white background The before and after images are shown below Now we will demonstrate a few other features of the software 1 Zi 10 Open Image1 tif Zoom in and Zoom out using the toolbar shortcut icons or the keyboard shortcut hot buttons Control and Control Change the Cursor for pixel sampling View Select Pixel Sampling Cursor 4 Flip the image horizontally Process Flip the Image Horizonal and vertically Process Flip the Image Vertical Convert to Negative Image Process Convert to Negative Image Filter Color To Grayscale Convert to Negative Image Rotate the image clockwise Process Rotate Image Clockwise Accept 90 default OK Print functions File Print Preview Active Image Close File Print Setup File Print toolbar shortcut icon Help Menu Display the About Shield Help About CMEIAS Color Segmentation display the user manual Help User Manual connect to the CCMEIAS website Help CMEIAS website display Color Segmentation page and also Contact Us page Thanks for watching this tutorial l l l Email me at cmeiasfd msu edu Thanks for watching this tutorial you have any questions E Frank Dazzo
26. ge with objects of background noise removed by this size filtration routine 29 h Smoothen Sel This filter smoothens the jagged edges of foreground objects by softening the color transition between their edge pixels and background pixels Since only the edge pixels of the foreground objects undergo change no detail is lost This filter is useful when cutting copying and pasting selections to create composite images i Sharpen Object Edges This filter applies a mild focus on foreground objects whose edge is somewhat blurred Its Sharpening algorithm operates by increasing the contrast of adjacent pixels only at the edges of foreground objects while preserving the overall smoothness of the image j Find Object Edges The Find Object Edges filter provides an interactive way to isolate a foreground object and erase its background and the inner regions Pixels on the edge of the object lose their color components derived from the background so they can blend with a new background The primary purpose of this filter is to delineate the outline edges of objects as a precursor for other image processing tasks For instance it can be used to quickly visualize which foreground objects are at the edge of the objects of interest and the extent that foreground objects are touching each other Fig 21 A second use is to produce a derivative image from which the foreground objects can be isolated copied and pasted into other applications Fig
27. gmented output image containing several erroneously included background pixels B red arrows which were efficiently removed by applying the color erosion filter C Caution Use these Dilation and Erosion operations carefully and conservatively Applying Color Dilation will merge very close objects together into one object and Color Erosion will NOT remove all of the added dilated pixels 21 f Fill Small Holes Occasionally under sampling of foreground object pixels in the original color image may produce a color segmented output image containing pixels of background color completely enclosed within some foreground objects The Fill Small Holes post processing feature is provided to correct this type of false dismissal error It automatically converts internal foreground pixels lost during color segmentation into the object s average color The program must have in memory the RGB value of at least one sampled foreground pixel in order to compute the mathematical morphology algorithm to perform this image processing task so before applying this feature to fill holes with a new batch of images you must first select at least one foreground object pixel To access this routine select Process Fill Small Holes and then specify the upper size limit of the holes to fill An example of its use is illustrated in Fig 11 where holes of different size within an object were filled using the hole size settings specified in pixels Objects
28. he sensitivity of the color segmentation function The color distance of this threshold setting represents the radii of all spheres in 3D color space of included colors whose center is the sampled pixel s A higher threshold value allows more distantly related colors to be included in the segmented image while lower threshold values narrow the range of colors that will be included The default value is arbitrarily set at 100 The dialog box that opens when this process is selected will accept color distance unit values from 1 to 400 Fig 6 Use of high color distance threshold values would aid color segmentation when the background and other colored pixels present in the image are significantly different from the foreground colored pixels of interest Conversely a low color distance threshold value would be used to more effectively segment images in which the RGB colors of interest are close in value to colors of background that need to be discarded Figures 7A 7B and 7C illustrate the color similarity tolerance function ska 65 100 gt 150 Fig 7A Image segmentations applied to a red region of a color spectrum with Color Tolerance Limit set at 65 100 and then 150 Note how an increase in color similarity tolerance setting allows a greater deviation in color from the sampled pixel to be included in the output image 18 Figs 7B 7C Use of the color similarity tolerance feature to achieve accurate color segment
29. ics of color hue saturation brightness intensity Fig 14 Hue is the color reflected from or transmitted through an object It is measured as a location on the standard color wheel expressed as a degree between 0 and 360 The position at 12 o clock is at 0 360 In common use hue is identified by the name of the color such as red orange or green aX Saturation sometimes called chroma is the strength or purity of the color Saturation represents the amount of gray in proportion to the hue measured as a percentage from 0 gray to 100 fully saturated On the standard color wheel saturation increases from the center to the edge a Intensity is the relative lightness or darkness of the color usually measured as a percentage from 0 black to 100 white 4 Convert to Pseudocolors Original Image w Pseudocolored Image Fig 15 Example of an RGB image converted to pseudocolors to enhance recognition of the object s edge In this example cells have a fluorescent halo that obscures their surface This option converts the active image into another with different colors that reflect the brightness intensity of the image pixels The program does this by first converting the image to a grayscale and then assigns an RGB color for each increment of gray level brightness This tool will work for both grayscale and RGB color images This process can help to recognize the periphery of foreground objects when they
30. ify the name and location of the saved text file from which the pertinent RGB range of training pixels is to be retrieved This function will clear out any existing sampled pixels before loading the new set from the file Additional sampled pixels can be added to the loaded pixel set and resaved The color segmentation process will still work even if some of the coordinate locations of the loaded pixel set are beyond the bounds of the image to which the set is being applied Nice CMEIAS shortcut tool m Discard Sampled Pixels Tl This command clears the current set of sampled pixels Use this function frequently for example before sampling pixels for color segmentation of a new image 5 FILTERS An image filter is a process that changes the shades and pixel colors of an image Filters are used to increase brightness and contrast and add various textures tones and special effects to an image 26 a Color to Grayscale The Color to Grayscale filter converts the active 24 bit RGB image into the corresponding 8 bit grayscale image Every pixel of the grayscale image has a brightness value ranging from O black to 255 white computed from the brightness values of its red r green g and blue b chromatic channels according to the following formula b 11 g 59 r 30 100 As an example Fig 18 illustrates the color to grayscale conversion of fluorescent yellow green bacterial cells to grayscale pixels with the corresp
31. in indistinctive boundaries that contrast gradually with the background This digital heterogeneity may not be noticeable when the image is viewed at 1 1 100 zoom but is obvious when magnified to view the color of individual pixels comprising the microbial objects Fig 2 Solving this challenging segmentation problem is crucial when any computer assisted microscopy application uses color information Fig 1 to extract ecologically relevant quantitative data especially at the resolution and spatial scale of individual microbial cells and their ecological niches within environmental samples Figure 2 Zoomed in detail of digital images showing variation in colored pixels comprising individual bacterial cells and the indistinct fluorescent halo surrounding their boundaries due to the bending of light as it passes through the cell The color stains and their corresponding RGB ranges within these cells are A FITC r92 r118 g198 g255 b0 B DTAF r116 r179 g166 g246 b167 b227 C crystal violet r62 r157 g0 b167 b227 D DAPI r1 r74 g49 g191 b157 b255 E rhodamine r124 r217 g0 b1 b8 Bar scales equal 0 5 um Image from Gross et al 2009 Most often color segmentation of microbial images is addressed by isolating the foreground object pixels with a single or narrow RGB color range and or splitting the color image into its individual RGB chromatic channels followed by thresholding the channel that contains the most intense signals fo
32. indicates the objects that were within the defined size bounds Fig 20C Selecting the colored buttons labeled Included Objects and Excluded Objects Fig 20B allows customization of the color scheme Choose pure black and white for 8 bit binary images B ox ID OK Range of abject size to include Cancel Range of object size to include Cancel Generate Preview Colors Generate Preview olores gt at Max Pixels sae Re Included Objects 30 Min Pixels Excluded Objects fag Min Pixels reset defaults Fig 20 Sequence of steps to use the Minimum Maximum Object Size Filter to remove background noise from an image A Color segmented output image containing regular rod shaped foreground objects and background noise of small spheres and a large irregular shaped object of the same color B Dialog box displayed when Generate Preview and Optimize Size is selected C Preview output image displayed after applying the size filter to the input image Steps B and C are repeated to optimize the size range of foreground objects The settings have been optimized for this example since the light blue foreground objects are discriminated from the orange objects of background noise whose size is outside the specified range D Dialog box to input the optimized size range when Apply Size Filter to Image is selected E Final output ima
33. int Setup command presents the standard Windows printer setup dialog box used to select the printer type of paper orientation printing quality number of copies and other information for future printing jobs i Print Active Image a Shortcut Ctrl P The Print Active Image command prints the image in the active window It supports any printer connected to your computer with the appropriate Windows compatible print driver installed j Recent Files The Recent Files menu feature lists the 9 most recently opened image files with their complete filename displayed for easily identification A single click of any image file listed will open it into the program workspace k Exit The Exit command will terminate the CMEIAS Color Segmentation application 15 2 EDIT a Undo 2 Shortcut Ctrl Z The Undo command removes the last process or filter change made to the active image and restores it to the previous condition Up to 8 iterations of the undo feature can be applied b Redo El Shortcut Ctrl Y The Redo command re performs the last undo operation again on the active image c Copy Shortcut Ctrl C The Copy command allows you to copy the active image to the Clipboard where it can then be transferred to any Window s compatible application that supports the paste command You can also use this feature to transfer the active image to a graphics package for adding annotations e g bar scale image label arrows etc
34. it can result in a loss of detail in the image 28 g Min Max Object Size Filter Generate Preview amp Optimize Size Apply Size Filter This powerful MinMax Object Size Filter removes objects larger than the user defined max size and smaller than the min size of foreground objects in the current image It is typically used to remove non foreground pixels from an RGB color segmentation output image or an 8 bit grayscale image after it has been thresholded to binary Both submenu options of this filter Generate Preview Optimize Size Apply Size Filter require the user to define the size range of objects to be INCLUDED in the new image by indicating the minimum and maximum pixel areas for all objects considered as foreground When applied the filter paints a user defined background color to all pixels of objects whose size is outside the specified area range while retaining all pixels of foreground objects at their input color within that same size range Fig 20A E Appling this filter to an image will not generate a new image So it is advisable to use it with a duplicate rather than the original image The Generate Preview submenu feature Fig 20B is used to optimize the MinMax Operation and recolor the image in accordance to what was inside and outside the given area bounds In the new preview image generated by default orange color denotes objects that were outside the size range of the given area bounds while a light blue color
35. ive non transferable right to use CMEIAS Color Segmentation Software for research or educational purposes You may not redistribute transfer rent lease sell lend sub license prepare derivative works decompile or reverse engineer this CMEIAS Color Segmentation Software without prior express written consent of MSU at the above address MSU retains title to CMEIAS Color Segmentation Software including without limitation the Software and Documentation End User agrees to use reasonable efforts to protect the Software and Documentation from unauthorized use reproduction distribution or publication All rights not specifically granted in this Agreement are reserved by MSU Warranty CMEIAS Color Segmentation Software and Documentation are provided as is MSU MAKES NO WARRANTY EXPRESS OR IMPLIED TO END USER OR TO ANY OTHER PERSON OR ENTITY SPECIFICALLY MSU MAKES NO WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE OF CMEIAS SOFTWARE OR DOCUMENTATION MSU WILL NOT BE LIABLE FOR SPECIAL INCIDENTAL CONSEQUENTIAL INDIRECT OR OTHER SIMILAR DAMAGES EVEN IF MSU OR ITS EMPLOYEES HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES IN NO EVENT WILL MSU LIABILITY FOR ANY DAMAGES TO END USER OR ANY PERSON EVER EXCEED THE FEE PAID FOR THE LICENSE TO USE THE SOFTWARE REGARDLESS OF THE FORM OF THE CLAIM General If any provision of this Agreement is unlawful void or for any reason unenforceable it shall be deemed severa
36. ive Image T The Duplicate command creates a new window displaying a copy of the current active image If an image stack is opened only the active top image will be duplicated Use this feature to compare the original image to processed images derived from it 14 e Close Active Image The Close command closes the currently active image window f Acquire Image The Acquire Image command is used to acquire digital images using a TWAIN interface e g digital camera flatbed scanner To use TWAIN you must install the TWAIN driver provided by the device manufacturer and the appropriate Source Manager for the scanning device you are using The Source Manager is a DLL file available at http www twain org index html Note that there are separate versions for different Windows operating systems so make sure that you install the correct one This Acquire Image menu feature has 2 choices 1 Select Twain Source Use this command to select the external Twain compliant device to acquire images When selected a window will open allowing you to specify which Twain compliant device installed on your computer to use 2 Acquire TWAIN Image When selected the device driver and software for the external Twain compliant device that you have selected will be opened allowing you to acquire the image g Print Preview Active Image This option displays the active image in a standard window before it is printed h Print Setup The Pr
37. l Sample Foreground Pixels 8v FR Apply Color Segmentation PY assunto Post Process If Needed Dilate Erode Fill Holes Min Max Size Grayscale Save Segmented Image Figure 3 Flow of steps to segment the foreground objects in color images using the CMEIAS Color Segmentation software application Image from Gross et al 2009 11 Figure 3 shows the general flow of steps used in this software application to segment foreground objects in color images with a complexity range typically encountered in microbial ecology studies First images are opened in the graphical user interface and examined to assess the similarity in color heterogeneity of the foreground objects and in their contrast to the background Six different cursor designs have been included to optimize the precision of this pixel sampling process for a wide range of images In addition the RGB values of the pixel under the cursor automatically display in the status bar to assist with this initial evaluation The information provided by this first step is used to set the color range tolerance on a scale that defines the range of color to be included in foreground pixels near each sampled pixel s location Next training pixels are interactively and carefully sampled from the objects of interest in each region of similar color within the active image The required number of pixels sampled varies depending on their color heterogeneity within the foreg
38. ning image s soso oaoa 33 b Segment the red fluorescent cells in Imagel tif 4 4 34 c Segment the Gram negative spiral bacteria in Image2 tif 36 d Segment the blue fluorescent cells stained with DAPI in Image3 tif 37 e Segment the red green and blue cells in Image4 tif 4 39 f Additional training not included in the AVtutorial 40 1 Segment the green fluorescent cellsinImagel tif 40 2 Segment the Gram positive rods inImageS tif 41 I CMEIAS Color Segmentation Michigan State University Software License Agreement By downloading and installing a copy of this CMEIAS Color Segmentation Software and Documentation you agree to the following terms Notification of Copyright CMEIAS Color Segmentation Software is a proprietary product of Michigan State University MSU and is protected by copyright laws and international treaty You as End User must treat CMEIAS like any other copyrighted materials Copyright laws prohibit making copies of the Software for any reason You may make copies of the Documentation for use with a licensed version of the Software however MSU notifications of copyright must be left intact If you have any questions concerning this agreement please contact the Copyright Licensing Office MSU East Lansing Michigan 48824 U S A 517 355 2186 or 517 432 4499 License MSU grants End User the royalty free non exclus
39. noisy background while still including the very thin spirochetes as intact foreground objects 2 Set the background color to white Process Segmented Image Background White 3 Set the color tolerance level to 100 Process Color Similarity Tolerance set at 100 OK 4 Discard any sampled pixels in memory il 5 Sample 5 7 pixels from cells near the borders around the entire image Alternatively load the previously saved set of RGB values for the sampled foreground pixels of the fluorescent bluish white Dapi stained cells for this image Process Load Sampled Pixels Select Image3pixels txt Open 6 Apply the color segmentation routine Process Apply Color Segmentation de 7 When using the Image3pixels txt sampled pixel file note that the upper left quadrat of the color segmented image output contains several cocci with internal holes painted with background white pixels Fill those holes within these cells using the Fill Small Holes process with the default hole size setting of 20 pixels Process Fill Small Holes OK The color of the pixels used to perform this process is the average RGB value for the entire cell 8 If necessary apply the dilate erode editing process to ensure that the pixels of the spirochete cells are continuous Process gt Apply Color Dilation Process Apply Color Erosion Pe ay Before Before gt After image After image 36 Segment the red green and blu
40. onding brightness as foreground Use this feature to convert final color segmented images to 8 bit grayscale for CMEIAS semi automated object analysis and classification Figure 18 Conversion of a 24 bit RGB image to the corresponding 8 bit grayscale image using the Color to Grayscale filter o a This feature is also used to prepare images for quantitative CMEIAS image analysis of object luminosity For example to measure Gfp gene expression the fluorescent image is first color segmented then converted to the 8 bit grayscale image then inverted to produce the negative image and finally quantitatively analyzed for grayscale brightness In this way the luminosity of the foreground objects is placed on a scale of 0 255 that is proportional to the brightness intensity of Gfp gene expression Gross et al 2009 b Brightness Threshold The Brightness Threshold tool converts a grayscale image to a binary image containing only pixels of pure black 0 and white 255 typically as a preface to further object analysis When selected it opens a dialog box to input a threshold value level 0 255 An automatic thresholding algorithm computes the default value An OK response converts the active grayscale image to binary whereby its pixels with a brightness of less than the input value are converted into 0 black and pixels with brightness above the input value are converted into 255 white If the active image is a 24
41. r the targets of interest while suppressing the intensity of the other channels This approach has variable degrees of success when applied to digitally pseudocolored monochrome images such as those acquired as a primary grayscale image using confocal laser scanning microscopy and then pseudocolor processed for specific fluorochromes Implementing other image processing routines such as dilation erosion Gaussian blur contrast manipulation spatial convolution masks c means clustering classification of pixels into predefined pseudochannel classes mean median filtering and measurement feature descriptors for object size and or shape filtration can sometimes help to minimize blurred object edges and complement color channel based image segmentation of microbes However combinations of these image processing routines rarely succeed in segmenting the 3 dimensional color space that accurately defines all foreground pixels of microbial targets of interest at all locations within complex true bitmap color images to analyze their size shape abundance and spatial location in situ In addition underlying assumptions e g RGB intensities of foreground object pixels are approximately equal to each other and greater than intensities of background objects are not always valid and the original true color intensities of the foreground objects are inevitably lost using these routines since they are typically applied to the whole image even when only selected area
42. re being processed The Save and Load Sampled Pixels features also provide a useful shortcut when rerunning the color segmentation process at different color threshold tolerances to optimize its crucial setting When displayed the fully segmented 24 bit RGB output image or its 8 bit grayscale image derivative can be copied to the Window s clipboard or saved directly as is B Menu Options a CMEIAS Color Segmentation File Edit wiew Process Filters Window Help Saya ees eRe eje 7 ele E E Zoomi 2 1 x 134w 30 RGB 255 255 255 Pixels Sampled 2 Time st 0 002 Image Size 200x120x24 Fig 5 Title bar main menu items toolbar shortcut icons and status bar of the CMEIAS Color Segmentation graphical user interface 1 FILE a Open Shortcut Ctrl 0 The Open command loads images to display at their 1 1 original size TIF BMP and PNG image file formats are supported Multiple images can be opened to form a stack but only the most recently opened image in the active window can be processed The time sec with 0 001 precision required to open the image will display in the status bar b Save i Shortcut Ctrl S The Save command saves the active image to its same filename Images can be saved in any of the supported image formats c Save AS The Save As command saves the active image to a different filename and or location Images can be saved in any of the supported image formats d Duplicate the Act
43. round objects and how isolated are those regions within the image Doing this pixel sampling task while viewing the cells in a zoom mode can be helpful The time required will depend on the size and complexity of the image number of sampled foreground pixels and whether the signal and noise contents of the currently active image are sufficiently similar to previously segmented images whose array of sampled training pixels had been previously saved Once these interactively trained inputs are registered the color segmentation algorithm is activated to analyze the image pixel by pixel using the color and spatial ranges specified by the user selected training pixels and the specified threshold value that defines their 3 dimensional color space to determine which pixels are to be included as foreground objects The run time to complete this automated algorithm depends on the size of the input image number of training pixels sampled and speed of the computer This computing time is reported in milliseconds in the status bar and typically takes less than 2 seconds using a PC with Pentium 4 and 3 00 GHz CPU to process images commonly included in microbial ecology studies After the pixel classification is completed the software automatically creates and displays a new color segmentation output image in which the pixels of foreground objects are retained in their original color and position and the non foreground pixels are painted either black or white a
44. s require them We sought to minimize these limitations by developing a more accurate efficient robust and versatile algorithm to semi automate the segmentation of multicolored microbes in digitized color images that also contain complex and usually noisy backgrounds and to implement this improved technology into a well documented and user friendly PC software application Earlier versions of our color segmentation algorithm are described in Reddy et al 2003 and Reddy et al 2004 In more recent work we improved on the segmentation algorithm to achieve 99 accuracy over a wider range of complex segmentation scenarios and describe the computer vision logic of our new system the accuracy of its significantly improved color segmentation algorithm sources of error and how they are addressed and examples of its application to solve various complex image processing challenges commonly encountered in color images acquired for quantitative microbial ecology studies Gross et al 2009 This free computing toolkit facilitates the integration of microbial ecology with cutting edge individual single cell microbiology at spatial scales directly relevant to the microorganisms themselves This system is a component of CMEIAS Center for Microbial Ecology Image Analysis System whose combined purpose is to strengthen microscopy based approaches for understanding microbial ecology at um spatial scales III General color segmentation protocol The
45. se ee 4 23 K Save SampledPIxels a e s amp s pm e a a e ews 24 k Load Sampled Pixels sios o E RAS 24 m Discard Sampled Pixels nono wee we 24 Si DULCES ai AAA nl E 24 a Color to Grayscale o c s aoi a acs ath a mo 25 D Brigntness Thresholds e goze E ER RA 25 AQUUSt AU Saturation as mes 2 oe A A A E 26 d Increase Intensity ao a a a a ee 26 e Decrease Intensity els a e Ste se ra a e E Rir 26 AgjustContrast ia e A AD e A A E E 26 e Min Max Object Size Filter sas rr RES 27 y SINO OLE ma demo ti Jane e DESP GS PAD Ae RD a a 28 Ix sharpen ODject dees sus sra pas E Md E EA 28 k Ema Object Ed CS sm sms dd od E RE DE ee Sk A E E 28 Embora da Dorada 29 Os WdoWa a e ra A ve AA AAA 30 di Cascade s m Giese eh Be tee eg She RGR A o es a ER E 30 Ds M x parada ie ae a a Oe Ads a 30 Er Cose IW dase ch thw ae a o o sd Ee ee 30 d Windows Lists a e seio A Bo ee ee ee es a Ow ae Ee a 30 de Me oa Bare Tah a DD a aa 30 a About CMEIAS Color Segmentation s soso oa ee ee 30 hb User Manual gs ata a a Aa Sid 30 Cy Help VODICS ve mg de e E a A A A tg 30 Gs CMEBIAS NM ebsm ss ms E E E E A A SE 31 We Reet MCCS ss sr na eo e Aub ee A RES SO he as A fee A ce 32 Appendix 1 Training Tutorial amp Images a aoso ee ee 33 a Apply these steps to each trai
46. segmentation Fig 9 20 SA ImagelAtf o amp s Figure 9 An insufficient number of training pixels were sampled from the original image A resulting in the color segmented output image containing several erroneously excluded foreground pixels B which were efficiently added back by applying the color dilation filter C e Apply Color Erosion ES The option applies an erosion filter to the active color segmented image resulting in a slight reduction in size of the foreground objects background size increases This filter operates by converting foreground pixels that have at least 1 background pixel neighbor to the background color It can be applied to RGB or grayscale images This and the Color Dilation filter are mainly applied to correct certain types of errors in the color segmented output image made by the Apply Color Segmentation routine For example use the color erosion filter to eliminate single isolated pixels representing background noise erroneously included as foreground in the color segmentation output image and background pixels classified as foreground at the edge of an object called a false alarm error during color segmentation Fig 10 Y ImagelAtf o amp es Copylofim o amp Ez YP Copy2 of Co 1 E P C Figure 10 An insufficient number of training pixels were sampled from the original image A resulting in the color se
47. t the discretion of the user to optimally contrast the noise free background Subsequent iterations of this sequence plus combinations of other image post processing features Figs 3 amp 4 can be applied further to refine the results of the output image and produce the final image segmentation desired IV Menu Structure A Open Save Save as Duplicate Close Active Image Acquire Image Select Twain Source Acquire Twain Image Print Image Preview 12 Overview _ Process Segmented Image Background gt White Color Similarity Tolerance Apply Color Segmentation Apply Color Dilation Apply Color Erosion Fill Small Holes Cursor 1 Color Models Cursor 2 Cursor 3 Cursor 4 Cursor 5 Cursor 6 Print Setup Print Active Image 1 Image tif 2 Image2 tif gt Split to RGB Split to YUV Split to HSI Convert to Pseudocolors Flip Image gt Horizontal Vertical Convert to Negative Image Rotate Image Clockwise Save Sampled Pixels Load Sampled Pixels Discard Sampled Pixels Filters Window Help About CMEIAS Color Seg User Manual Help Topics CMEIAS Website Cascade Tile Close All Color to Grayscale Brightness Threshold Adjust Hue Saturation Increase Intensity Decrease Intensity Add Contrast ari Figure 4 The menu structure of CMEIAS Color Segmentation including preprocessing and post processing functionalities added to compliment the color segmentation algorithm thereby
48. tial impact of the study itself Microscopy of Microbes Using Color Epi Z a Autofluorescence Fluorescence Ue A Transmitted Engineered Natural E Autofluorescence Autofluorescence Brightfield re 7 EPT Differential Gram Stain Metabolic Activity Membrane Integrity Figure 1 Hierarchical organization of various types of epifluorescence and transmitted light microscopy that utilize the discriminating power of color information to reveal significant characteristics of microorganisms Abbreviations are FISH fluorescent in situ hybridization CTC 5 cyano 2 3 ditolyl tetrazolium chloride FITC fluorescein isothiocyanate Gfp green fluorescent protein DTAF 5 4 6 dichlorotriazinyl aminofluorescein DAPI 4 6 diamidino 2 phenylindole dihydrochloride ELF PO4 and SYTO BC are commercial trademarks Image from Gross et al 2009 The challenge of color segmentation is how to separate foreground pixels from background along fine delineations of color and location within the complex image The underlying problem is that microbial objects of interest in high definition digital color images are commonly represented by pixels with heterogeneous brightness ranges of red green and blue RGB that most often also include colored pixels of background at similar locations and the pixels often have shallow gradients of brightness transition at cell borders resulting
49. tolerance level to 100 Process Color Similarity Tolerance set at 100 OK Discard any sampled pixels in memory a Load the image5pixel txt file of saved sampled pixels Process Load Sampled Pixels image5pixel txt OK Apply the color segmentation routine Process Apply Color Segmentation amp If necessary fill small holes using the 20 pixel default setting Process Fill Small Holes 20 pixels OK
50. w bottom middle using crucial adjustments in color tolerance before applying color segmentation 19 c Apply Color Segmentation amp This is the most important function of this software Objects of interest e g bacteria within color images will consist of pixels with widely varying RGB values Fig 2 and therefore it will be difficult or impossible to segment an image to contain just the foreground object pixels of interest based on sampling a single pixel value CMEIAS Color Segmentation applies a nearest neighbor approach to classify all the pixels within the active image and isolate those foreground pixels of interest based on the representative range of RGB values and their spatial proximity to the training pixels sampled Once completed the active image is converted into the color segmented output image with foreground objects in a noise free user specified black or white background Fig 8 Before activating this tool some pixels must be sampled click the mouse left button to sample the pixel under the cursor maxim um of 200 on the image that will approximate the RGB range of foreground objects of interest e g bacteria Status messages display when 190 and 200 pixels have been sampled See Gross et al 2009 for full documentation of the computer vision logic of this color segmentation algorithm in our new system measurements of its accuracy 99 when tested on ground truth data from a wide variety of color images of mi
51. x provides the text and the Before After image illustrations for the CMEIAS Color Segmentation Training Demo CmeiasColorSegmentation wmv distributed with the software The audio visual demo runs in Windows Media Player a To begin apply the following same steps to each training image 1 Start the program and maximize its workspace on your computer monitor 2 Opening the image File Open 3 Make a duplicate working copy of the image File Duplicate the Active Image T 4 Close the original image File Close Active Image 5 Adjust the position and size of the copied image window so it fully displays at 1 1 zoom 14 6 Discard any sampled pixels in memory Process Discard Sampled Pixels al 7 Evaluate the range of colors for the pixels of foreground cells by viewing their RGB values in the status bar as you move the cursor over the cells 8 Follow the steps indicated below to segment the foreground bacteria of interest in the training images provided 35 Before gt After image a E na ii b Segment the red fluorescent cells in Image1 tif 1 2 Open Image1 tif Set the background color for the output image to black Process Segmented Image Background Black While viewing the RGB values for pixels of red cells note that they include intensities in both red and green chromatic channels but no values of blue Set the Color Similarity Tolerance setting to 65 pixels Process
52. xels First select Filters MinMax Object Size Filter Generate Preview and Optimize Size accept default settings of 30 min 1620 max OK This will display a preview image with the foreground objects included within the specified size range colored as light blue and all other objects whose size lies outside the filter range colored as brownish orange default color settings that the user can change In this case the default settings of 30 pixels minimum to 1 620 pixels maximum work OK in discriminating the foreground objects Gram negative spiral bacteria colored aqua in the size filter preview image from background noise of red pixel halos around the purple Gram positive rods colored brownish orange in the size filter preview image After approving the optimized settings indicated by the preview image apply the size filter to the color segmented output image again Filters MinMax Object Size Filter Apply Size Filter to Image accept optimized settings OK Voila This procedure produces the final segmented output image with the red spiral bacteria accurately segmented in the noise free white background 37 Before gt After image Md o 6 Op o A o o y Y a O O A c Segment the blue fluorescent cells stained with DAPI in Image3 tif 1 Open Image3 tif The main challenge in segmenting this image is to fine tune the balance of color similarity tolerance so it adequately removes the
Download Pdf Manuals
Related Search