sofware user`s manual - Ciencias de la Computación e Inteligencia
Contents
1. 1 8964 Displacement x x Figure 13 Display the best HR reconstruction EA outopt Eile Edit View Graphics Debug Desktop Window Help CEEL te ta Stack Base hs AI plot outopt betak Min Max L 4 9222 4 3636 5 6988 Value 3 8420e 08 4 9222e 1 3873 0 0 0956 0 0618 0 03 0 0686 0 6 2729 4 1989 0 4 62729 0 5 4883 0 4143 3 06 5 4883 54 54 lt 154 double gt 0 0017 7 5642e 06 aae 7 5642 lt 54 double 27 6710 51 2420 51 2124 51 2124 51 2124 lt 1x54 double 0 0 0 0 0 51 2420 51 2420 51 2420 0 1253 1 0022 lt 84x84 double gt 51 51 51 4 1989 6 2729 3 0683 5 4883 0 0956 0 0686 0 6 2729 4 1989 0 4 0 5 4883 0 4143 3 06 0 0 0956 0 0618 0 03 mated HR image using the method in 2 14 Simulated X LR Generator Number of LR images 4 Magnification Factor 2 Average X Warp Random MY Mask size odd 3 Displacement x 0 4 8741 3 1707 1 668 Imam Less SNR 3 Rk Super Resolution l i t Reference image 1 Magnification Factor 2 Registration Parameters R Known true lt MSE Deconvolution PSNR Lambda 0 1 0 0 SAR 0 75 SR Methods E using combination Norm L1 SAR z Figure 15 Coefficient value for the prior model combination when2. Iteration Number MSE 1 3949e 005 48 554593 dB Cancel 35 Running MODE RESET LR Generator Number of LR images 5 Magnification Factor 2 Average M Warp Random M Mask size odd 3 Displacement x v 0 4 8925 7 0633 0 970 1 at ow ean a R Super Resolution Reference image 1 Registration Parameters Magnification Factor Known true lt Deconvolution SR Methods SR using TV Prior Results ee Iteration Number MSE 1 103e 005 49 574408 dB 112 Completed 0 4 8925 7 0633 0 9701 1 8964 Displacement x lt Figure 12 Execution of the SR method finalized 13 Low Resolution Image 5 Size 42 X 42 Figure 14 Stored parameter in the structure outopt when saving the esti MODE RESET Number of LR images 5 Magnification Factor 2 gt Warp Random LR Generator Average Mask size odd 3 Displacement X v 0 4 8925 7 0633 0 970 Over hsm Super Resolution Reference image 1 Magnification Factor 2 Registration Parameters Known true lt High Resolution Image Results Iteration Number MSE 7 94e 006 PSNR 51 001802 dB True Warp Completed 0 4 8925 7 0633 0 9701
3. 0 0 Noise l Super Resolution Reference image 1 Magnification Factor 2 limulated Mode Registered a2 7 iteration Number 19 Deconvolution Completed Size of Mask 3 SR Methods SR using combination Norm L1 SAR Lambda 0 1 0 0 SAR 0 7 Estimated Warp Diopiacement x zj 0 16542 1 4832 1 6695 Figure 18 End of execution of method in 3 in Real Mode with A 0 7 at iteration 19 16 Figure 19 Starting execution of method in 5 Figure 20 Definition of filter in execution of method in 5 17 PD d Versioon743Public SRSoftware FiltersDir x Buscar FiltersDir Organizar v Nueva carpeta 3 v J Nombre Fecha de modifica Tipo Tan E Dxmat 29 06 2015 17 49 MATLAB Data E DxDySAR FilterList mat 29 06 2015 17 49 MATLAB Data Z Domat 29 06 2015 17 49 MATLAB Data E Dxy mat 29 06 2015 17 49 MATLAB Data S Dy mat 29 06 201517 49 MATLAB Data E Dyxmat 29 06 2015 17 49 MATLAB Data FN4_FilterList mat 29 06 2015 17 49 MATLAB Data E FNS FilterList mat 29 06 201517 49 MATLAB Data d E mean mat 29 06 2015 17 49 MATLAB Data 9 E NE mst 29 06 2015 17 49 MATLAB Data 1 Ss Figure 21 Definition of filter in execution of method in 5 18 Filter List List Management BG Load List of Filters cag j Versioon743Public gt SRSoftware FiltersDir lt la BuscarFiltersDir Organizar v Nueva
4. Estimated registration sx 1 66946 sy 1 83849 theta 0 264915 Estimating the hyperparameters avg DxSigma 0 00198266 avg DySigma 0 00198266 k 1 e 0 0576815 xOx 0 traceBkSigma 0 189113 traceOSigma 0 k 2 e 0 0497142 xOx 0 traceBkSigma 0 145384 traceOSigma 0 k 3 e 0 0431147 xOx 0 traceBkSigma 0 145948 traceOSigma 0 k 4 e 0 0365582 xOx 0 traceBkSigma 0 146958 traceOSigma 0 betak 7147 66 9041 6 9330 24 9612 23 alpha_h 3 69758 alpha_v 3 86702 xconv 0 0361037 Figure 23 Content of the log file after running the method in in real mode with A 0 7 in iteration 2 20
5. acti vated Now the Low Resolution Image area allows to save the generated observations The observations are saved by typing a generic name in the file manager window For instance if the user generated 5 observations and set the generic name lenag png then the application creates the image files lenag_1 png lenag_2 png lenag_8 png lenag_4 png y lenag_d png Now we can apply the SR methods implemented in the software to the generated observations Notice that Reference image Registration pa rameters and SR Methods are the only areas enabled in simulated mode The subarea Registration parameters allows the user to choose between using the true registration parameters or estimate them True registration parameters are used by default The next step consists of selecting the desired SR method to execute as shown in figure At this point there are 5 active areas Low Resolution Image Mode Low Resolution Image 0 Size High Resolution Image m MODE RESET Simulated LR Generator 4 Blur Without blurring Warp Random tonne Buscaren images amp ef Ey S RL RL RL t Sitios recientes lena4_15 png lena4_16 png lena4_17 png lena4_18 png High Resolution Image Simula f 7 lenag80_1 png lenag80_2 png Nombre lenag80 png L be Tipo ipa gif tif png Figure 4 GUI appearance after clicking the Open button High Resolu
6. method in 3 or 4 is selected Number of Observatior 1 Size 0 X 0 Hi S MODE RESET L E LR Generator Number of LR images 4 Magnification Factor 2 Blur Without blurring warp Random v Displacement X S b EZ Super Resolution Reference image z Magnification Factor igh Resolution Simulated Mode ee Registration Parameters Results MSE Deconvolution PSNR SR Methods cwe v Figure 16 Real Mode 15 10 Ne SS Se eS ae vations 4 Size 42 X 42 MODE High Resolution Image Size 84 X 84 Real LR Generator r of LR imag 4 Magnificat t 2 Blur Without blurring Warp Random Displacement x 0 0o o o m Noise war SNR l LR r Super Resolution Reference image 1 Magnification Factor 2 tion Image Simulated Mode Registered r Results S F iteration Number 4 Deconvolution ce aiken 3 l Running Average ize of Mask Cancel ll SR Methods SR using combination Norm L1 SAR lt Lambda 0 1 0 0 SAR 07 Open SR Figure 17 Execution of the method in 3 in Real mode with A 0 7 Size 42 X 42 MODE High Resolution Image Size 84 X 84 E lt LR Generator R image 4 Maanificatio actor 2 Blur Without blurring Warp Random Displacement x 0 0
7. registration sx 1 66946 sy 1 83849 theta 0 264915 Estimated registration sx 1 66946 sy 1 83849 theta 0 264915 Estimating the hyperparameters avg DxSigma 0 00115548 avg DySigma 0 00115548 k 1 e 0 522118 xOx 0 traceBkSigma 0 113152 traceOSigma 0 k 2 e 0 410507 xOx 0 traceBkSigma 0 0871786 traceOSigma 0 k 3 e 0 401887 xOx 0 traceBkSigma 0 0880384 traceOSigma 0 k 4 e 0 391035 xOx 0 traceBkSigma 0 0886754 traceOSigma 0 betak 2776 77 3544 41 3600 55 3677 22 alpha_h 4 7704 alpha_v 4 84304 xconv 0 0388807 Iteration 2 Finished calculating Sigma registration terms Running PCG to estimate the image PCG returned the solution at 81 iterations flag 3 Estimating the registration RRR RRR RRR RRR RRR RRR RRR RRR RRR RRR Image 2 Initial registration sx 1 65418 sy 0 843637 theta 9 43549 Previous registration sx 1 65418 sy 0 843637 theta 9 43549 Estimated registration sx 1 65418 sy 0 843637 theta 9 43549 RR RR RR RRR RH RRR RRR RRR RRRRRRRRRRRRE Image 3 Initial registration sx 1 48317 sy 0 214256 theta 9 14108 Previous registration sx 1 48317 sy 0 214256 theta 9 14108 Estimated registration sx 1 48317 sy 0 214256 theta 9 14108 RR RRR RRR RRR RH RRR RRR RRRRRRRRRRRRE Image 4 Initial registration sx 1 66946 sy 1 83849 theta 0 264915 Previous registration sx 1 66946 sy 1 83849 theta 0 264915
8. Superresolution software manual S Villena M Vega D Babacan J Mateos R Molina and A K Katsaggelos Version 1 0 Contact information Rafael Molina Departamento de Ciencias de la Computaci n e I A E T S de Ingenier a Inform tica y de Telecomunicaci n Universidad de Granada 18071 Granada Spain e mail rmsQdecsai ugr es This manual describes the use of the software application implementing the SR algorithms developed in 4 5 as well as other superresolution methods developed by the authors 6 The application has been developed in MATLAB 7 18 R2011b including the graphic user interface see fig ure 1 The developed software e Allows the execution of SR methods in real and simulate modes e Evaluates the following SR methods Bicubic interpolation The TV prior based method proposed in 6 The method described in the algorithm 3 1 of 1 which utilizes a SAR prior model The algorithm proposed in 2 with the 1 norm of the horizontal and vertical gradients as the image prior model The method proposed in 3 based on a SAR and 1 norm model combination with registration parameter estimation The method described in the work 4 which combines SAR and TV prior models with the estimation of the registration parame ters The method proposed in 5 where a prior distribution based on a general combination of spatially adaptive or non stationary image filters is used with the estimation of a
9. anada and Leonidas Spinoulas and Michail Iliadis Ph D students at Northwestern University for testing the software This work has been supported by the Comision Nacional de Ciencia y Tecnologia under contract TIC2010 15137 Note Please report bugs to Rafael Molina rms decsai ugr es License This program is free software you can redistribute it and or modify it under the terms of the GNU General Public License as published by the Free Software Foundation either version 3 of the License or at your option any later version This program is distributed in the hope that it will be useful but WITH OUT ANY WARRANTY without even the implied warranty of MER CHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE See the GNU General Public License for more details You should have received a copy of the GNU General Public License along with this program If not see lt http www gnu org licenses gt References 1 S Villena Superresoluci n y Reconstrucci n Bayesiana de Im genes a Partir de Im genes de Baja Resoluci n Rotadas y Desplazadas Com 10 LT Low Resolution Image 0 Size 0 X 0 High MODE RESET v LR Generator Number of LR images 4 Magnification Factor 2 Customized Warp Random Results SR Methods Bicubic Figure 8 Setting the warp binacion de Modelos PhD th
10. carpeta B il a Nombre R Fecha de modifica Tipo Tama o T DxDySAR_FilterList mat 29 06 2015 17 49 MATLAB Data 1 Gre T FN4_FilterList mat 29 06 2015 17 49 MATLAB Data 1 T FNS_FilterList mat 29 06 2015 17 49 MATLAB Data 1 T P1_FilterList mat 29 06 2015 17 49 MATLAB Data 1 p2 FilterList mat 29 06 2015 17 49 MATLAB Data 1 P12_FilterList mat 29 06 2015 17 49 MATLAB Data 1 N 9 Figure 22 Load a list of filter in execution of method in 5 19 Running SR with optimal distributions Calculate initial average image betak 957 763 1168 29 1152 24 1138 75 alpha_h 9 32744 alpha_v 9 47627 Iteration 1 Finished calculating Sigma registration terms Running PCG to estimate the image PCG returned the solution at 88 iterations flag 1 Estimating the registration RRR RR RRR RRR RRR RR RRR RRR KK KK KKK KK KK K X Image 2 Initial registration sx 1 65418 sy 0 843637 theta 9 43549 Previous registration sx 1 65418 sy 0 843637 theta 9 43549 Estimated registration sx 1 65418 sy 0 843637 theta 9 43549 RRR RRR RRR RR RRR RRR RR RRR RRR RRR RRR RRR Image 3 Initial registration sx 1 48317 sy 0 214256 theta 9 14108 Previous registration sx 1 48317 sy 0 214256 theta 9 14108 Estimated registration sx 1 48317 sy 0 214256 theta 9 14108 RRR RR RRR RRR RRR RRR RRR RRR RRR RRR RRR RRR REE Image 4 Initial registration sx 1 66946 sy 1 83849 theta 0 264915 Previous
11. er Resolution panel is ac tivated allowing the user to set the magnification factor set to 2 in each direction by default and the blurring kernel without blurring by default in the Deconvolution subpanel As for the simulated mode the user selects the SR method and presses the SR button to run it The appearance of the panels High Resolution Image and Results are shown in figure 17 Once the method has finished the registration parameters are shown accessing iteration number you can retrieve previous iterations reconstruction and the results can be saved by pressing the Save button see figure 18 3 Files output during the execution The application displays the execution status in the GUI but it also saves log files in the directory tempSR which contains the information of all the parameters estimated in each iteration An example of the content of this file for the real mode execution depicted in figure 18 is shown in figure 23 B Matrix Reader e rex Input Matrix Size 9 RH ER 9 HRS R E E Save Matrix Load Matrix O O OOO O O O G G G G oO gt oOo o G e O O gt oOo oOo oO oOo G O O O OOOO oO oo O OOO O O O OOOO O gt O O OOOO O Ji 2 4 5 6 7 m LK Figure 7 This matrix allows the user to define the blur kernel in the Cus tomized mode Acknowledgments We want to thanks to Miguel Tallon Ph D student at University of Gr
12. esis Universidad de Granada December 2011 S Villena M Vega R Molina and A K Katsaggelos Bayesian super resolution image reconstruction using an l1 prior In 6th International Symposium on Image and Signal Processing and Analysis ISPA 2009 Best paper award Image Processing and Analysis Track pages 152 157 2009 S Villena M Vega D Babacan R Molina and A Katsaggelos Bayesian combination of sparse and non sparse priors in image superres olution Submitted to Digital Signal Processing 2012 S Villena M Vega D Babacan R Molina and A Katsaggelos Us ing the Kullback Leibler divergence to combine image priors in super resolution image reconstruction In IEEE International Conference on Image Processing pages 809 812 Hong Kong China September 2010 S Villena M Vega R Molina and A K Katsaggelos A non stationary image prior combination in super resolution Digital Signal Processing 32 1 10 September 2014 S D Babacan R Molina and A K Katsaggelos Variational Bayesian super resolution IEEE Transactions on Image Processing 20 4 984 999 2011 11 Low Resolution Image 4 Size 42 X 42 MODE RESET Number of LR images 4 Magnification Factor 2 LR Generator Average X Warp Random Mask size odd 3 Displacement x v 0 4 5614 0 5618 0 S SS Nse S Ove SNR 30 allin m Super Re
13. ess can so be monitored Low Resolution Images 0 Size 0 X 0 S High Resolution Image Siz 0 4 0 MODE S 5 Super Resolution Reference image Magnification Factor 3 High Resolution Image Simulated Mode Registered 7 S Results SR Methods Figure 1 GUI appearance Numbers 1 to 7 inside the red circles indicate the interface interaction areas 7 Results In simulated mode this area shows error measures of the HR reconstruction and the value of the real and estimated parame ters if applicable In real mode it shows the estimated registration parameters The only GUI element active at the beginning is the area Mode The user must then select one of the two available execution modes real or simulated 2 2 Execution Mode The execution mode can be selected by clicking on the MODE RESET pull down menu see figure 2 Note that the Real mode is chosen by default Depending on the selected mode the GUI activates the fields where to provide the information needed to run the selected SR method Each execution mode is described in a separate section since the selected mode determines the application workflow Selecting the execution mode at any time involves resetting the applica tion to the selected mode All the loaded images and obtained results will be lost and all the parameters will be reset to their default values Dont Ts Number of Observatior 1 Size 32 X 32 High Res
14. ll parameters If you use this software to evaluate any of the described methods please cite the corresponding papers The papers can be downloaded for research purposed from http decsai ugr es vip e Saves the results HR images and registration parameters e Allows the user to stop the execution at any time with the possibility of saving all the estimated parameters as well as the estimated HR image We assume here that a digital camera captures a sequence of L low resolution images of size N pixels In the spatial domain the process to capture the k th image YL can be expressed using matrix vector notation as Vk AH C s x Ip 1 where x represents a digitalization of the real scene with a resolution of PN pixels with P gt 1 representing the magnification factor C s is a PN x PN matrix representing the motion of the k th image in the sequence with respect to the reference image with sz 0 Ck dg the motion parameters rotation angle and horizontal and vertical displacement H isa PN x PN matrix representing the blur A is the N x PN integration and subsampling matrix and nz represents the noise This model is used in the software to simulate the low resolution observed images The simulated mode allows to evaluate the quality of the resulting HR images and the error in the estimated parameters for each method see the experiment sections in chapters 4 6 of IUT In contrast the real mode execution allows to r
15. olution Image Real da Simulated 4 MODE RESET LR Generator Blur Without blurring Warp Random Displacement x Noise var 3 SNR LR Open Show Images End Super Resolution Reference image Magnification Factor High Resolution Image Simulated Mode Registered Results Deconvolution Blur Without blurring Figure 2 GUI appearance when the area MODE RESET is active 2 2 1 Simulated Mode When the simulated mode is selected the Open button in the area High Resolution Image Simulated Mode is activated see figure 3 When the Open button is pressed the file manager is launched this allows to select the HR image to be used in the experiments see figure For instance if we select the image lenagS0 png the image is displayed and the LR Generator panel becomes available see figure 5 This panel shows all the required information needed to generate the LR images using Eq 1 All fields have been initialized to default values that can be modified at any time The blurring function is defined by the selected kernel as shown in figure 6 Note that for each type of blur the values of the blur matrix coefficients are requested by the application They are initialized with a default value in order to facilitate the user input When the Customized option for the blur is selected a new window is di
16. on 7 13 0 564 R2011b and it has been tested to work only in this version 2 User interaction with the application The main steps to run the application are 1 Start the application 2 Select the Real or Simulated 3 Fill in the initial information 4 Select the SR method to execute 5 Execution monitoring and saving options Let us describe each step in details 2 1 Running the application To run the MATLAB application change your current directory to the application folder and run the program SuperResolutionv1 m In a few seconds the GUI depicted in figure 1 will appear The GUI is divided in the following interactive areas 1 Mode Reset Allows the user to select between simulated or real execution mode and to reset the program 2 Low Resolution Image this area shows the utilized simulated or real observations depending on the current execution mode 3 High Resolution Image Simulated Mode Used only in simu lated mode It is utilized to load and previsualize the HR image from which the LR observations will be generated 4 LR Generator Only active in simulated mode This panel al lows to introduce the required information to generate LR observations from the image loaded in the previous area 5 Super Resolution Allows to select the SR method to be applied to the LR observations 6 High Resolution Image This area displays the resulting HR image on each iteration The reconstruction proc
17. outopt see figure 14 fa SuperResolutionv1 Low Resolution Image 0 Size 0 X 0 eae a High Resolution Image Size 0 Simulated X Number of LR images 4 Magnification Factor 2 LR Generator Average Warp Random Mask size odd 3 Displacement x High Resolution Image Simulated Mode m SR Methods Bicubic Figure 5 GUI appearance after the initial HR image is loaded When any SR method using prior model combination is selected the combination coefficient A is shown as depicted in figure The com bination coefficient in the range 0 to 1 is displayed on a new field The default value 0 means that the algorithm is exclusively applying the SAR prior model When the SR method in 5 entitled SR using Mix Filters is selected the window in figure is presented Pressing the Editor button the figure appears and you can define new filters and save them Pressing the New button the figure appears and you can create a new list of filters in the FulterDir directory from previously created filter files Care must be taken of selecting only filter files and not list of filters those ending on FilterList mat Pressing the Load button you can load previously saved list of filters as shown in figure 22 Filters and filter lists can be stored in the directory FltersDir When the filter list has been completed
18. solution Reference image 1 Magnification Factor Registration Parameters Known true Deconvolution Figure 9 Observations generation S mulated hdj LR Generator Number of LR images 5 Magnification Factor 2 4 Average Warp Random Mask size odd 3 Displacement x v 0 4 8925 7 0633 0 970 C Noise ee eer a Ea Super Resolution Reference image Magnification Factor 2 Parameters Results Known true S a Deconvolution SR Methods Bicubic SR using TV Prior SR using Norm L1 Prior SR using SAR Prior SR using combination Norm L1 SAR SR using combination TV SAR Figure 10 Available SR methods 12 Low Resolution Image 5 Size 42 X 42 Low Resolution Image 5 Size 42 X 42 MODE RESET LR Generator Number of LR images Average gt Warp Random 5 Magnification Factor 2 Mask size odd 3 Displacement x v J 0 4 8925 7 0633 0 970 Nase a I ans a SNR 30 a Super Resolution Registration Parameters s SR using TV Prior Reference image 1 Magnification Factor High Resolution Image Results
19. splayed in order to introduce the blur coefficients see figure Addi tionally one can select either the Save option to save the current kernel or Load to load a new one The registration parameters can be defined in three forms Warp Ran dom Customized and Example depending on whether you want to generate them randomly introduce the parameter manually rotation angle and vertical and horizontal displacements for each observation or using the parameters in the experiments described in chapters 4 6 of L respectively see figure Sl Finally the user must set the zero mean Gaussian noise level which Low Resolution Image 0 Size 0 X 0 High Resolution Image Size 0 X 0 High Resolution Image Simulated Mode Figure 3 GUI appearance when the area Mode is activated to Simulated Note that the button Open is now active can be defined in two forms introducing the noise variance or by setting the signal to noise ratio SNR both in exclusive buttons Depending on the selected option the numeric field allows to introduce either the noise variance or the SNR value The SNR mode is active by default with a value of 30 dB The LR button is available once all the required information to gen erate the observations has been provided see figure 9 Notice how the areas Low Resolution Image and Super Resolution have been
20. tion Image Simulated Mode LR Generator and Su per Resolution That means that you can modify the fields of any of these areas and generate new observations with the possibility of saving the previous ones The selected SR method is executed by pressing the SR button The areas High Resolution Image and Results are now enabled as displayed in figure 11 On each iteration the HR reconstruction and its quality metrics are displayed The execution can be stopped at any time by pressing the Cancel button When the method stops due to convergence or cancellation the application shows the true registration parameters and if applicable the estimated ones see figure 12 Additionally the application allows the user to display the best recon struction in the PSNR sense by pressing the Best Results button that could not coincide with the solution obtained in the last iteration as seen in figure If the button is pressed again the HR reconstruction in the last iteration is shown Both reconstructions can be saved pressing the Save button In order to request the file name to save the HR reconstruction the file manager is launched Let us point out that the application saves the HR reconstruction in the last iteration with the given name the best recon struction in the PSNR sense with the same name but ending in Best and it also saves a file with the same name but ending in Dat mat containing the estimated parameters in a data structure named
21. un the methods with real datasets This mode has been utilized in the real experiments in chapter 6 of I Depending on the selected mode and before running the selected method the user has to provide the information required by the method Addition ally if the simulated mode is selected the user has to introduce the necessary information to generate the observed low resolution image set according to Eq 1 that is the original HR image the number of observation to gen erate the blur the additive Gaussian noise variance and the subsampling factor In real mode the user has to provide the LR image set the assumed or previously estimated blur and the desired upsampling factor Once the initial information has been filled in the user has to choose the SR method and run it Note that when selecting any of the proposed methods that use model combination it is also necessary to introduce the combination parameter A in the range 0 1 The user can intuitively navigate through the GUI and complete the initial information needed by the SR method Also the user can access the reconstructed HR image and registration parameters on each iteration in case they were estimated The access and workflow through the developed GUI is detailed next 1 Installation The application does not need to be installed To access the application just uncompress the provided package The application was developed in MATLAB versi
22. you must press the Finish the Filter Selection to Mix button to optionally save the list and run the method 2 2 2 Real Mode The Number of Observations panel and Open button are activated on Real mode see figure 16 When pressing the Open button the file manager is launched requesting for the first observed image file name The rest of options Show Images fi SuperResolutionv Low Resolution Image 0 Size 0 X 0 High MODE RESET a eS L LR Generator Number of LR images 4 Magnification Factor 2 Average M Warp Random Average Gaussian 0 0 0 Disk l gt Displacement x Customized L S lt l Sar Sa var SNR 30 LR onow images Save Super Resolution Reference image Magnification Factor Registration Parameters Results Deconvolution SR Methods Bicubic Figure 6 Blur kernel selection and End are available once the first observation is introduced Adding new observations is as simple as repeating the procedure until the last observation is opened and the End button is pressed he user can also load all the observed images at once by selecting a mat file containing a single variable of size rows x columns x n observations that stores all the images The show images button displays the set of images currently loaded When the End button is pressed the Sup
Download Pdf Manuals
Related Search