Performing Iterative Non-blind Deconvolution with BiaQIm
Contents
1. J Tadrous How to Deconvolve with BiaQIm www bialith com Page 4 Using a prior file To use a prior file you must tick the Use prior file gt box and then load the prior file to be used This may be loaded by simple drag n drop into this area but the prior must be of the appropriate type and dimensions and if it is of a raw data type there must be an external header qih file present in the same directory for drag n drop to succeed Alternatively bring up a File Open browser by a single left click to the Prior button and load the prior file that way The Use Constant prior at value tick box and edit box lets you specify a plain image which is constructed internally to use as prior where all pixels in that image have the same intensity value namely the value you enter in the edit box The Use MEAN as prior also constructs a plain prior image for you this time with all its pixels set to the mean average intensity value of all pixels in the input image Step 3 3 Next go to the PSF tab and drag n drop the PSF file into the depressed rectangular space in the upper part of this window labelled with PSF File in green figure 11 5 b This will load the PSF file and the PSF height and width will automatically be displayed ensure that the Normalise box is ticked this is the default state of the box An alternative way to load a PSF file is to bring up a File Open browser by a single le2. Myrheim amp Rue maximum entropy methods the Other tab changes its appearance to look like that shown in figure 11 7 b Here the Alpha value edit box is replaced by a Typical Noise Variance edit box and the constraints tick boxes are not available Instead you have low and high thresholds on Copyright 2008 by Dr P J Tadrous How to Deconvolve with BiaQIm www bialith com Page 6 the weights file which correspond to the entries in the settings file called VarThreshold_LH and Weight_set_LoH1 described above in the section on settings files Another new setting is the Lowest value in p edit box corresponding to Mi ni mum __ Value which you must set to a number between 0 and 1 usually close to 0 Notice also that there are now two depressed regions on the right of this tab The top one which is used for adding a weights file in the other methods is now used to add an optional variance file in a similar way you must tick the Use Var File box to actually use any variance file loaded into here The lower depressed area operates in a similar manner but is used to add a weights mask you must tick the Use Mask box to actually use any weights mask file loaded into here Loading a variance file or weights mask is done via drag n drop or file browser brought up by a single left click to the Variance file or Weights Mask buttons al D econy olu tior a ee fb on Figure 11 7 a The Oth
3. a new process causes some of these settings to return to default status Step 5 Run the process Before you actually run it tick the Load result box in the Process tab of BiaQIm and enter a name for your resulting image in the edit box for Im such as LWMoon fit see figure 11 9 Now left click once on the START button Assuming you have the deconvolve exe program on your PATH or in the CWD the process will start in a new DOS window that will open up automatically figure 11 10 You will be able to monitor the progress of deconvolution as the current iteration and residual will be written out to the DOS prompt at each iteration and the best result presented in brackets next to it The best minimum residual is usually the current iteration s value so this will change with every iteration provided the solution is converging If the solution begins to diverge then this minimum value will no longer be the current iteration s value and you will be able to see this in the DOS prompt Copyright 2008 by Dr P J Tadrous How to Deconvolve with BiaQIm www bialith com Page 7 QI Process 10 Configurator ici RM oon jo x IHPUT S OUTPUT S In Out Display Mask Measure Process Z START HEMNE FILL V amp LUE rl BMoon 23 fit i 4 Figure 11 8 Set the options in the IO Configurator window as shown in the figure highlighted with ellipses Do the following in order
4. button in the top half of this tab the depressed area labelled Settings File in green This will bring up a File Save dialogue box for you to select a name The file will be saved in the appropriate format depending on which deconvolution algorithm you selected back at step 3 1 You may also load a settings file from disc This can be done via the Load button which brings up a File Open dialogue box or you may simply drag n drop a settings file into the depressed area Either way if you load a settings file and it is of a valid format then all the drop down menus tick boxes and edit boxes will automatically be set according to the settings in that file This settings file load and save facility is a useful way of saving your settings from a previous deconvolution experiment so you can continue at a later date Before moving on to step 4 here is some information about settings not used in the current example but which have not yet been described Going back to the Other tab illustrated in figure 11 6 a you will notice a depressed area in the top right of the window This provides a facility to use a weights file a raw array of doubles of same dimensions as the input image described in the earlier sections dealing with the settings files If you have such a file and you also have an external qih header file present in the same directory you may simply drag n drop the weights file into this depressed area and
5. if you use those algorithms For our present example we will use constraints to ensure a non negative solution because the Landweber algorithm is a least squares method that does not have an inherent non negativity constraint unlike Lucy Richardson Super Resolution or maximum entropy methods which are all non negatively constrained by design Tick the Apply Constraints box and set the Lower box to 0 and the Higher box to some high number well above the maximum intensity level you would expect in the true solution in this example I use 1024 The other available options for this tab will be described later but are not used in the current example Step 3 5 Finally select the Admin tab and in the section dealing with the log file settings in bottom half tick the box labelled Based on OUTPUT file name to ensure that the log file generated by the program has a name that is based on the resulting deconvolved image You could alternatively use one of the other options or un tick all these three boxes and select a custom file name by bringing up a browser dialogue box via a single left click to the Log file button and typing or selecting a name to use clicking Save on this browser dialogue just saves the file name you selected to BiaQIm it does not save anything to disc If you want to you may now save all the settings you have made over the last five steps by a single left click to the Save
6. the values of the residual at this iteration and in brackets the iteration and value of the residual for the iteration that had the minimum residual value If the solution is uniformly converging then this minimum value will always be the current iteration s value Deconvolved image lteratrve Mon Blind Dr P J Tadrous ae sas FILL VALUES a fal Load result LiwMoon fit Ima Figure 11 11 When finished the DOS box will close the deconvolved image will automatically be loaded and displayed in the main display window of BiaQlm because you ticked the load result box earlier and you will notice two new files in your folder One is the resulting image file WMoon fit and the other is the log file LWMoon isv Copyright 2008 by Dr P J Tadrous How to Deconvolve with BiaQIm www bialith com Page 9 Making sense of the results The result of performing this process is a deconvolved image called LWMoon fit and a log file called LW Moon tsv figure 11 11 The command line and settings The command line constructed for you by BiaQIm can be seen on the Cmd tab and will be this its Al e fits 12 D WorkArea DCTest BMoon128 fit fits WI NDOWS T deconvolve ex 8 1 Qldcs txt 8 12 LWMoon fitC e mp From this it can be seen that BiaQIm has created a temporary settings file for you called QlIdcs txt in CA WINDOWS Temp You may read or copy this file but as noted in step 3 5 above
7. How to Deconvolve with BiaQIm www bialith com Page 1 Performing Iterative Non blind Deconvolution with BiaQIlm Introduction This is an extract from the BiaQIm User s Manual More detail is given in chapter 11 of that manual where the background to the programs used in this example is explained including basic information on how to use BiaQIm the format of the deconvolution settings file and other step by step examples Chapters and figures referred to in this text refer to the full BiaQIm User s manual which you can download from www bialith com together with all the software The download installation package for BiaQIm includes a folder called DCTest which contains the image files shown in this example so you can work along with it and do the deconvolution yourself Please remember that any image files you use with BiaQIm must not contain spaces in their name and must not be located on a path that contains spaces such as My Documents or anything in Documents and Settings This is because the external processes such as deconvolution are DOS based programs that cannot handle such path names If you install the BiaQIm package in the default location then you should be able to perform this example without problem Bear in mind also that your Windows system must be running in Windows Classic mode if you have XP or Vista otherwise BiaQIm will not start See the notes on the Software Download page of www bialith co
8. er tab settings for the Lucy Richardson algorithm b The Other tab settings for the Myrheim amp Rue maximum entropy algorithms Another point of difference in the setting between Myrheim amp Rue maximum entropy methods and the other algorithms is in the Pre Process tab For Myrheim amp Rue maximum entropy methods you will not get the options of using the SOLVE border treatment neither can you set the prior image to a constant value of your choice or to the mean value of all the pixels in the input image as you may with the other algorithms Step 4 Configure the inputs and outputs Once you have finished defining all the settings click the OK button on the settings dialogue box to confirm your settings choices this is essential because the settings do not take effect until you click OK This window will close and the IO Configurator window will appear figure 11 8 Detail about the IO Configurator is given in chapter 8 but for the purposes of this example just set the output drop down list to read Im and ensure that the input and output data types are both set to FITS as shown in figure 11 8 Now close the window Note you do not have to close this window for the I O settings to take effect In many cases you may want to leave it open somewhere to ensure you always have the right input output settings because these can change due to various actions you may take subsequently and selecting
9. erative deconvolution See figure 11 3 Copyright 2008 by Dr P J Tadrous How to Deconvolve with BiaQIim www bialith com Page 2 Hork reasOCT est b Moon ALl File liszt S tretc H Stretch Limits Ignore Dr PJ P Ue wee inis Lower Upper Pedestal Central Outer ESE EA Licepabileet I 7 nolo 104 Height ize Type i Load Boon 28 fi o ra SutoLo erly Valed 2103 width Skip PSF17 fit i ae So Bytes wari Dy 2g 1 object selected a fez ot d Figure 11 2 Drag the image BMoon fit and drop it into the QI logo area of BiaQlm The image will be displayed in the main display window OI Process Menu LUTION Enhancement PSF Estimation Transforms Ol La Click this button to bring up the process menu Filters Restoration Segmentation E Batch Modifier Extn Binary Other X Cancel Figure 11 3 A Select the Process tab B Click the Choose button circled on the Process tab of BiaQlm to bring up the process menu C Move your mouse over the word Restoration on the left to bring up the restoration sub menu then move across to this sub menu and select the Iterative Non Blind process as shown The process menu will disappear and the deconvolution settings dialogue will open up Copyright 2008 by Dr P J Tadrous How to Deconvolve with BiaQIm www bialith com Page 3 Step 3 Set the parameters via the settings dialogue box The deconvol
10. f iterations and set this to 4096 you need to type this in the box These settings are shown in figure 11 4 Note that the settings options displayed may change depending on the convolution algorithm you select For example if you selected Maxent 2 Myrheim Rue you would get four convergence criteria boxes not 3 and these will be labelled according to the convergence criteria available for that algorithm Other changes will occur in other tabs too according to the specific algorithm chosen at this stage QI Deconvolution ei BMoon P Fa Set these three parameters as E shown then move on to the next tab labelled Pre Process DUT est 0 Ww ork reasDCT est T BMoonl 28 fit Figure 11 4 Select the appropriate settings for the deconvolution process Algorithm convolution type and convergence criteria Step 3 2 Now select the next tab along called Pre Process figure 11 5 a and tick the Solve box at bottom right This ensures that the SOLVE boundary method is used to avoid ringing from the sharp contrasts at the boundaries The prior option should be left at Use input image as prior because this usually gives the best results when using the SOLVE method The other options are self explanatory once you have read and understood the sections above on the settings files Although not used in the present example the method for using a custom prior file 1s explained below Copyright 2008 by Dr P
11. fd Boon fel E 1 Tick Load result 2 Type the name of the output image in the edit box for Im2 have chosen to call it LWMoon fit 3 Left click once the START button These are shown highlighted by magenta ellipses Bialllm DCTest BMoor 23 it TE ESE lt Figure 11 9 Before finally running the process first tick the Load result box so you will see the solution image as soon as the deconvolution program terminates and type a name for the output image in the edit box labelled Im2 In this example use the name LWMoon fit Now click the START button and the program should begin to run in a new DOS window that will pop up for this purpose Copyright 2008 by Dr P J Tadrous How to Deconvolve with BiaQIm www bialith com Page 8 MS 2 deconvolve Parameter value not in allowed range Bad command or file name De WorkAreasDCTest gt Deconvoluve v2 8 Copyright c 1994 2667 Dr Paul J Tadrous This program is Freeware but is MOT in the Public Domain For License Agreement visit wuw bialith com Itn 268 13 8242 CMin 13 8242 at Itn 268 _ Figure 11 10 Once running the program will progress in a DOS window and BiaQlm will be unresponsive until the DOS window closes The BiaQlm interface may not refresh if obscured during this time This is normal With deconvolution the output at the DOS prompt is updated with each iteration and tells you the current iteration
12. ft click to the PSF file button and load it this way Ol Deconvolution Ol Deconvolution lterative Non Blind weiner 7 LSG lterative Non Blind weiner 7 LSG Algorithm Pre Pracess Other Admin Algorithm Pre Process PSF I Use prior file gt Fion PSF file Ea Use input image as prior DO AWorkstea DCTestPSF17 fi E Use Constant prior at value E Use MEAN as prior ea E Momalise aI E Blank weights border M Solve a Figure 11 5 Select the appropriate settings for the deconvolution process a Preprocesses priors and image border treatments make sure you tick the SOLVE box as shown and b PSF settings If you simply drag n drop the PSF file into the sunken area labelled PSF File then it will automatically be loaded and the height and width displayed Otherwise use the PSF file button to bring up a File Open browser and load the PSF that way Ol Deconvolution Ol Deconvolution Iterative Non Blind weiner 7 LSO lterative Non Blind weiner 7 LSO Algorithm Pre Process PSF Uther Algorithm Pre Process PSF Other Admin i Use Wot File Weights file Apply Constraints i Flue preserving E Based on INPUT file name fd Based on OUTPUT file name E NONE generic a Figure 11 6 Select the appropriate settings for the deconvolution process a Other settings various constants and constraints weights files and masks and b Administrative settings input and
13. its file name and path will be recorded You can also do this via a File Open browser by a single left click to the Weights button However recording the weights file name in this way is not sufficient for it to be used in the deconvolution For that you must also tick the Use Wet File box The controls described in those tab are the same for the following deconvolution algorithms Landweber Van Cittert and Maximum entropy of Gull amp Daniell apart from the renaming of alpha to gamma in the latter For the Lucy Richardson method the Other tab changes its appearance to look like that shown in figure 11 7 a In this case we do not have an alpha or gamma setting but instead there is a Background value edit box Use this to specify the background grey level for a more accurate Lucy Richardson deconvolution This background level is the lowest grey level anywhere in the true solution for astronomical images this can usually be set at zero or the dark level if the images are not calibrated For many brightfield images this will be non zero However if you do not know what it should be you can usually get away with zero because no intensity level can be negative in any real world image that has been properly captured and calibrated This Background value setting is absent from the SuperResolution deconvolution algorithm version of the settings because it is not used in that algorithm For the
14. m for instructions on how to make your system change to Windows Classic mode A step by step example This step by step guide shows you how to perform Landweber deconvolution of a 128X128 pixel image in FITS format called BMoon fit by a 17x17 pixel PSF called PSFI17 fit which is also in FITS format These are illustrated in figure 11 1 and both have pixels of double precision floating point numbers It is important to ensure that both image and PSF have no negative valued pixels This is a very specific example for the purposes of illustrating the general method Only a few of the many various options are used in this example Figure 11 1 The input image left and PSF right The PSF is contrast stretched for the purposes of display Step 1 Load the input blurred image into BiaQIm This can be done by simple drag n drop as described in chapter 2 see the section entitled Loading an Image in Auto Load Mode and figure 2 4 See figure 11 2 Step 2 Select the deconvolution external process in BiaQIm How to do this was described in chapter 8 To recap first go to the Process tab of BiaQIm then click the Choose button and the process menu will appear Move your mouse over the word Restoration on the left to display the restoration submenu then move your mouse across into the restoration submenu and left click once on Iterative Non Blind This will bring up the settings dialogue box for non blind it
15. movie or scroll through it With regards to the data about the residuals at these intermediate point they are stored in the log file in tsv format so you can plot a graph of the convergence behaviour using any graphing package that can handle tsv input you may need to copy paste this part of the log file into your graphing package s datasheet spreadsheet rather than import the log file whole Copyright 2008 by Dr P J Tadrous
16. output files other than image related files These include the ability to save the settings file that BiaQlm will make for you using all your settings as well as load settings from a settings file you have on disk It also include the log file options If you have a settings file on disc then you can simply drag n drop it into the sunken area labelled Settings File and it will automatically set all the options in the dialogue boxes tick boxes and drop down menus in all the tabs according to the information contained in that settings file you can also do this by clicking the Load button and using a File Open browser Copyright 2008 by Dr P J Tadrous How to Deconvolve with BiaQIm www bialith com Page 5 Step 3 4 Now go to the next tab along labelled Other and ensure that the Alpha value text edit box contains the number 0 999 This is appropriate for most Landweber or Van Cittert deconvolutions but if you use maximum entropy of Gull amp Daniell method then this box will change to being called Gamma value and you should generally set a lower value e g 0 01 or 0 005 The higher the gamma the faster the convergence but the maximum entropy of Gull amp Daniell method will fail if this value is too large it means that you have assumed there is too little noise in the image For Lucy Richardson or Super resolution methods there is no alpha or gamma constant and so this box and its label will not be shown
17. ution settings dialogue box has a main tab section at the very top to select between the settings options for Iterative Non Blind deconvolution and the non iterative Weiner LSQ settings Because you selected Iterative Non Blind from the process menu in step 2 the deconvolution settings dialogue box will automatically open with the Iterative Non Blind tab selected for you This has 5 sub tabs of its own which we will work through one by one In the following these will just be called tabs and the settings for each of these 5 tabs are described in steps 3 1 to 3 5 below Step 3 1 The first tab figure 11 4 allows you to select the deconvolution algorithm and type of convolution used in the algorithm and boundary treatment in spatial convolution cases using two drop down menus You then have a series of text edit boxes to set the values for the convergence criteria the CCRIT values discussed in the section on settings files above Only 3 of these boxes represent the actual convergence criteria the two boxes on the right are for setting whether and how often you want intermediate results to be output this feature will not be used in the present example so just ignore these boxes for now In this example set the Algorithm to Landweber LSQ and leave the Convolution at the default Fourier The only convergence criterion to set in this example is the Max itern maximum number o
18. ve least squares deconvolution described in the second step by step example given in this chapter The log file and outputting intermediate results The log file generated is a plain ASCII text file containing a record of the deconvolution settings used In this record a value of 1 for a setting means that setting was not used in the deconvolution whilst a value of 1 or O in response to a question means YES or NO respectively The log file also records the input image and PSF and the time it took to do the deconvolution The log file also records the residuals data at each iteration that is output as an intermediate result to disc during the deconvolution In the current example the settings were such as to not allow output of intermediate results so only the final value of the residual is recorded in the settings file However if we repeat the procedure but this time at step 3 1 we set the number 2 in the Iterations edit box and also in the log edit box under Output current estimate every in the bottom half of the Algorithm tab of the deconvolution settings dialogue box see figure 11 4 we will get a record of the residual and an output of the current estimate as an image at the second iteration the A the 8 16 32 and so on This will be accompanied by a qfl file listing all the intermediate solution images that have been output so we can load this into BiaQIm and play the sequence as a
19. you can also save your own copy of this file via the Admin tab of the deconvolution settings dialogue box The deconvolved image The deconvolved image was produced by 4096 iterations of a constrained Landweber algorithm Figure 11 12 compares this result with the result using the Maximum entropy method of Myrheim and Rue with a typical noise variance of 0 01 and only 1024 iterations Boundary artefact problems appear greater in the latter but this is not necessarily a reflection of the technique s performance in general and it is inappropriate to go into details of deconvolution theory in this user s manual Now that you are familiar with the controls you can experiment with various options a b cka d Figure 11 12 Comparison of two deconvolution methods on the same input image a and PSF b Shows the result of the constrained Landweber method using 4096 iterations with an alpha of 0 999 and SOLVE border treatment Note that this solution was still converging when stopped so further iterations could produce a higher degree of deconvolution but you would probably have to double the number of iterations each time to notice an improvement c Shows the result of 1024 iterations of Myrheim amp Rue s maximum entropy with a typical noise variance of 0 01 and blanking of the weights border pixels Both show a similar degree of deconvolution but boundary artefacts appear greater than in b d Shows the result of one step non iterati
Download Pdf Manuals
Related Search