Published on June 3, 2016
slide 1: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 225 Image Restoration Using a Combination of Blind and Non-Blind Deconvolution Techniques Bassel Marhaba 1 Mourad Zribi 2 Wassim Khoder 3 12 Université du Littoral Côte d’Opale Maison de la recherche Blaise Pascal Laboratoire dInformatique Signal et Image de la Côte dOpale LISIC- EA 4491 50 Rue Ferdinand Buisson BP 719 62228 Calais Cedex France. 3 Université Libanais Faculté de sciences économiques et de gestion branche Tripoli-Liban nord Abstract — One of the important implementations in image-processing field is the image restoration. Image restoration deals with the recovery of an original image from a degraded image using a mathematical model of degradation and restoration for image. Image restoration is becoming more and more important in the image-processing field and it is very important in many applications like medical satellite and photography. In spite of the various existing solutions available to image restoration there is always a need for more efficient methods. In this paper several restoration and deconvolution techniques experimented and tested we used both blind and non-blind techniques. Then we propose a combination between blind and non-blind techniques in order to improve the quality of the restored image. Several types of noise are added to the image after it has been blurred. We have tested the behavior of the different filters and techniques in removing each type of noise. The evaluation of the filters behaviors and the conclusion are done based on various metrics like PSNR MSE RMSE and IEF. Keywords — Image processing Image restoration blind and non-blind techniques Noise Metrics. I. INTRODUCTION The restoration of the image is an area that also deals with improving the appearance of an image 13. However unlike image enhancement which is subjective the image restoration is objective in the sense that restoration techniques tend to be based on mathematical or probabilistic models of image degradation. In order to obtain a better restoration technique it is necessary to study and compare the details of the existing restoration filters and then to develop a more powerful filter which can fulfill the desire to have a cleaner image after removing the noise from it and achieve a powerful solution for the issues facing image restoration filters. In this paper one image will be restored using several restoration techniques after being degraded by being blurred and noise added. Several types of noise will be used in the degradation process. Noise in the image is that degradation of an image signal caused by an external disturbance when the image sent from one place to another place by satellite wireless or cable network. There are many types of images noise in this paper we will use the most common four types: 1- Salt Pepper noise which known as shot noise impulse noise or Spike noise. Its appearance is randomly scattered white or black or both pixel over the image. 2- Gaussian noise which can caused by random fluctuations in the signal it is modelled by random values added to an image. This noise has a probability density function pdf of the normal distribution. It is also known as Gaussian distribution. 3- Speckle noise it can be modelled by random values multiplied by pixel values of an image. 4- Poisson noise. Individual photon detections can be treated as independent events that follow a random temporal distribution. As a result photon counting is a classic Poisson process 2426.In section 2 we will declare image restoration steps provide a brief description of the blur function degradation model and restoration model. In section 3 the restoration techniques are described. In section 4 a proposed method will be illustrate and explained. Section 5 experimental results are shown. Finally section 6 we will give our conclusions. II. IMAGE RESTORATION STEPS The term Image Restoration means to restore an image from the degraded condition into a clear and restored condition. In other words it process is the degraded images which suffers from a blur and a noise in order to produce the output clear image. 2.1 PSF function Blur occurs commonly by a point spread function PSF. Image restoration techniques are divided into two classes according to prior knowledge about the PSF: Blind Image Restoration: The blind image restoration method allows the original images to be rebuilt from degraded images even with no knowledge or few knowledge about PSF. An example of the blind image restoration methods is the Blind Image Deconvolution BID 1. slide 2: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 226 Non-Blind Image Restoration: In this type of restoration a prior knowledge of the PSF that blurred the original image is occurred. This will help the restoration technique in rebuilding the degraded image. Deconvolution using Lucy Richardson Algorithm DLR Deconvolution using Weiner Filter DWF Deconvolution using Regularized Filter DRF are Non Blind Algorithms1. 2.2 Image Degradation Model In degradation model the image is blurred using degradation function H and then the noise is been added. The image degradation process can be modeled by 1 1: gxy Hxy. fxy+ nxy 1 where Hxy gxy fxy and nxy represent respectively the degradation function the observed or degraded imagethe original image or input image and the additive noise respectively. The function H represents a convolution matrix that models the blurring that many imaging systems introduce. For example the function H can model camera defocus motion blurs imperfections of the lenses all. 2.3 Image Restoration Model In the restoration model the image gxy was degraded is built back by the restoration filters.The restoration process is implemented by inversing the degradation process by removing the blur factor and the additive noise. We obtain an estimate of the original image after the restoration. The closer of the restored image fxy to the original image the more efficient is the filter. III. RESTORATION TECHNIQUES Different restoration techniques has been proposed in the restoration domain 24. In our paper we choose mostly common techniques to examine and test it by using different noise types applied on two different images format. 3.1 Non-Blind Deconvolution Techniques The non-blind techniques is that techniques which requires a prior knowledge about the blur function in order to process the restore operation. In our paper we offer the following techniques: 3.1.1 Lucy Richardson Algorithm The Richardson–Lucy deconvolution algorithm which is also named as Lucy Richardson Deconvolution LRD is a famous technique in the field of image restoration 22. Initially Leon Lucy and William Richardson derived it on the basis of the Bayes’s theorem in the early 1970’s 8. This method is categorized as non-blind deconvolution as it needs to know the PSF used to blur the image. It is also an iterative procedure. The pixels in the observed image are represented as in 2: 2 where d i is the observed value at the i th position of the pixel p ij is the PSF it represent the fraction of the light that comes from the true location j which has been spotted at position i u j represent the concealed image pixel value at the j th position. Our main goal is to calculate the most likely u j with the existence of the observed d j and the already known PSF p ij as following: 3 where 4 Lucy Richardson Deconvolution is easy to implement and it preserves edges as it is a nonlinear method. Specifically a big problem we face here is the noise amplification. The main issue for all maximum likelihood techniques which attempt to fit the data as closely as possible. When performing too much L-R iterations on an image that may contained an extended object such as a galaxy the extended emission usually develops a “Speckled” appearance 10. slide 3: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 227 3.1.2 Regularized Filter Regularized filter is one of the non-blind convolution family i.e. it de-blur an image with a prior knowledge of the blur function that blurred the image. This filter is considered an approximation for the Weiner filter and it result with a close result to that of the Weiner filter. Anyway the regularized filter need less information about the blurred function in order to restore the image. Regularized filtering is used in effective way when a few information is known about the additive noise. The regularized filter uses constrained least square algorithm to restore the noisy and blurred image. Regularized filter is usually classified as a discrete laplacian filter 1.Regularized filter is easy to implement and needs less information about the blurred function. However it has to have prior information about the blur function. 3.1.3 Weiner Filter Considered a linear filter Weiner filter is also considered as non-blind deconvolution it removes the noise from the degraded image with a prior knowledge about the PSF 1. At the same time it eliminates the additive noise and inverts the blur effect. Weiner filter implements the deconvolution technique with the means of high pass filter - inverse filter - accompanied with a compression operation - low pass filtering- to remove the noise. It compares with an estimation of the desired noiseless image. The process of the Weiner filter is to input the degraded image to the filter the output restored image by means of the filter is obtained by 5: 5 where and represent respectively the original image the output or the estimated image the noise and the Weiner filter’s response. We can use wide window in order to eliminate the Speckle noise to preclude the blurring of the edges we can use small window. One of the main disadvantages of the Weiner filter is the mandatory of the knowledge the power spectra of the ungraded image and the noise. In the case of randomly noise it is hard to estimate a typical restoration for the image. 3.2 Blind Deconvolution Techniques Unlike the non-blind deconvolution techniques the blind deconvolution technique does not require any prior knowledge of the blur function in order to process the restore operation. The several techniques we choose to test and examine are listed below. 3.2.1 Mean Filter The Mean filter or in other words the Average filter is linear class windowed filter that is used to restore images from noise. The filter is kind of a low pass filter. The main idea of the filter is to deal with each pixel of the noisy image as follows: It takes the pixel and the surrounded neighborhood pixels according to the window size has been specified then sums all the values and divide by the number of the elements. This is the average value at last the filter replaces the old pixel with the new average value and so on until all the pixels in the noisy image are replaced by the average value. The process now is completed we have the filtered or resultant image. This operation is considered by6 and fig.3: 6 Mean filtering is a simple to build and easy to implement. On the opposite any undesired value of one pixel can strongly affect the mean value of all neighborhood pixels. The filter will replace incorrect values for the edge pixel which will yield to a blurry image 3. 3.2.2 Median Filter The technique of the median filter is similar to that for the mean filter but in median filter 3 we don’t calculate the mean value the filter arrange the values of the pixels in ascending order within the window and then choose the median value to replace the tested image. The 7 below describes the work of the Median filter: 7 slide 4: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 228 where and are respectively the restored image the noisy image and the window. Since the Median chooses one of the pixels value that already exist that means it produces no new values. Implementation of the Median filter is very easy. The Median filter is less sensitive than the Mean to extreme values outliers median removes these extreme values in more efficiency way. The Median filter works well almost only with the salt and pepper noise. It is not effective with other kinds of noise 3. 3.2.3 Wavelet Deconvolution The wavelet method is used widely in image processing fields’ such as image compression and in image restoration 312. Unlike conventional Fourier transform wavelet transforms based on small waves called wavelets. Wavelets which means the little waves such as Haar Daubechies Morlet etc. Wavelets are functions that are concentrated in frequency domain and in time domain surrounding a fixed point. The wavelet transform is designed in such a manner to give a reasonable frequency resolution with low frequency component which are the average intensity values of the image and good temporal resolution with high frequency components which are the edges of the image. We can summary the process of restoration an image with wavelet deconvolution technique into three main stations wavelet transform or decomposition 3threshold 13 and finally the noise removing 14 Wavelet transform it is computationally very fast also it is easy to perform. On other hand It is shift sensitive because input-signal shifts generate unpredictable changes in DWT coefficients and it lacks the phase information that accurately describes non-stationary signal behavior 314. 3.2.4 Bilateral Filter The Bilateral filter 15 is proposed by Tomasi and Manduchi it is a nonlinear filter and is used to reduce impulse noise from images. Bilateral filtering smooths images and preserves the sharp features of edges with the help of a nonlinear combination of nearby image values. This method is non-iterative and simple. The Bilateral filter kernel w b is the product of two sub-kernels Gray-level sub-kernel Distance sub-kernel 5. The Gray level sub-kernel is given by: 8 where and is the standard deviation of . The distance sub-kernel is defined by: 9 where and is the standard deviation of In order to reduce the noise this kernel must slide throughout the noisy image and after filtering the estimated output is given by 10: 10 The filter has been used for many applications such as texture removal dynamic range compression and photograph enhancement. The main advantage of bilateral filter is that it can remove high density of noise from the images which other filters cannot remove. Bilateral filter is not effective with impulse noise like Salt Pepper noise. Also it replaces the noisy and not noisy image pixels with filtered value and the resultant images are smoothed but not sharpen. 3.2.5 Adaptive local Filter Adaptive local filter 2 is also one of the famous filters that applied on the degraded image. The restoration technique in this filter depends on two statistical measurements mean and variance with a specific n m window: slide 5: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 229 11 where are respectively the local variance of the local region the local mean the variance of overall noise the pixel value at the position xy and the restored value. The Adaptive local filter is simple to design and fast. In general it has weak response due to its slow convergence. 3.2.6 Blind Image Deconvolution In Blind Image Deconvolution BID unlike the Lucy Richardson Weiner and Regularized techniques the restoration techniques process the degraded image without a previous known of the blurring function. This restoration technique works primarily to estimate the blurring function PSF secondly it acquires the degraded image using the estimated PSF that was done at the first step. Technically this operation can be executed in either ways iterated or non-iterated. In the iterated process the estimated PSF would get better more and more with each iteration and then the restored image would be acquired from the degraded image using the estimated PSF. In the non-iteration process the PSF will be obtained by one application of the algorithm which based on extract information. Then they obtained PSF will be used to restore the image from the degraded image. The main target of the blind image restoration technique is to estimate the blur function and the original image 1. An advantage of BID is that there is no need to know previously about the blur function PSF any way the BID is effective at low noise intensities. IV. PROPOSED METHOD Our method is to combine an image restored from a non-blind deconvolution with the same image restored by blind deconvolution in order to improve the quality of the restored image. We will use a combination method to combine a two of resultant images and obtain an image that is better this method called image fusion 19. Image fusion has several types such as the high pass filtering which is the classic method. Other modern methods exist such as: fusion based on laplacian pyramid uniform rational filter bank and discreet wavelet transform. We will implement the combination using fusion based on discreet wavelet transform 20. The process of the fusion method is illustrated in fig. 1 The effective work in the wavelet based image fusion is to combine the coefficients in other words is to find the most convenient way to integrate the coefficients in such a way to have the best quality of the fused image. There are many ways to achieve this goal the simplest way is to calculate the average of the coefficients to be integrated 27. FIGURE 1 PROPOSED METHOD FOR IMAGE RESTORATION DIAGRAM Restored image 1 Combination method using fusion technique Fused image Restored image 2 Non-Blind Deconvolution Blind Deconvolution Degraded Image Original Image Degradation Model slide 6: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 230 V. RESULTS 5.1 Evaluation of the different techniques To evaluate the performance of our deconvolution methods we will apply it the bateau.jpg image. Four types of noise and blurring will be applied to the image before restoration. The evaluation of the performance of the restoration methods will be made based on the following metrics: RMSE the root of the MSE PSNR IEF and the execution time. MSE PSNR and IEF are given by 5: The Mean Square Error MSE is defined by 12: col rows j i f j i f MSE 2 ˆ 12 where j i f and ˆ j i f denote the intensity of j i th pixel of the original and filtered images. The Peak Signal to Noise Ratio PSNR is giving by13: MSE PSNR 255 255 log 10 10 13 The Image Enhancement Factor IEF is giving by 14: 2 2 ˆ i j i j j i f j i f j i f j i g IEF 14 where j i g is the intensity of j i th pixel of the degraded image. Our work has several dimensions in order to examine the several restoration techniques and declare the difference of the behavior between it we choose to apply several types of noise on our image deeper we will apply the different deconvolution methods on jpg image format. Salt Pepper noise will be applied at density of 0.06 Gaussian noise will have 0 mean and variance of 0.06 Speckle noise with 0 mean and variance of 0.04 and poisson noice with mean of 10. On other dimension we will examine the behavior of the restoration techniques with presence of blur only noise only and blur with noise. We will apply the blur only noise only and blur plus noise cases on the image bateau.jpg for fusion technique we will apply it on the image bateau.jpg only. The blur H or PSFfunction we will apply in our work is a Gaussian lowpass filter of size 5 with standard deviation of 5. We used MATLAB ® 2015 to obtain our metric results and the resulted images. 5.2 Results of the image restoration The bateau.jpg image is a gray image with size of 256 256 65536 bytes and uint8 class shown in fig.2. We will use this image in our three cases first the blur only second the noise only and third the blur plus noise. The four types of noise we mentioned in the previous section will affect the image and then it will be denoised. After that a fusion technique will applied in the sake of improvement the restoration result. Original image Blurred image only Salt Pepper only Gaussian only Speckle only Poisson Blur+ Salt Pepper Blur+ Gaussian Blur+Speckle Blur+Poisson FIGURE 2: ORIGINAL AND DEGRADED IMAGES slide 7: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 231 5.2.1 Case of Blur only In this section we will test the behavior of the deconvolution techniques in restoring the image with the blur only and no noise existing. The blur process is done with the PSF function which is a Gaussianlowpass filter of size 5 with standard deviation of 5.Then after the deconvolution of the blurred image with the deconvolution techniques we will choose the best result from the blind deconvolution techniques and the best result from the non-blind deconvolution techniques and propose the combination method of the two images in order to get a restored image with a better quality. In fig.2 we show the blurred image. TABLE 1 RESULTS OF DECONVOLUTION TECHNIQUES WITH BLUR ONLY EXISTING Filter Mean 33 Median 33 Weiner Lucy Richardson Bilateral RMSE 20.0799 19.9276 16.1429 13.8786 20.9403 PSNR 22.0756 22.1889 24.0266 25.1146 21.7111 IEF 0.8758 0.9857 1.8597 1.4006 0.8053 Restored image TABLE 2 RESULTS OF DECONVOLUTION TECHNIQUES WITH BLUR ONLY EXISTING Filter Adaptive Wavelet Regularized BID Fusion Technique RMSE 20.1879 19.41 17.037 21.2309 14.2 PSNR 22.029 22.3703 23.503 21.5914 25.92 IEF 58.5364 0.0722 1.2167 0.9259 1.72 Restored image In the Tables 1 and 2 we illustrate the results of de-blurring the blurred bateau.jpg image and the resulted fusion technique. We can notice that Lucy-Richardson deconvolution is the best over the others non-blind with PSNR of 25 in the other hand Wavelet deconvolution has the best result over the other blind deconvolution techniques with PSNR of 22.We combine two images using the wavelet based fusion method first of lucy-richardson deconvolution and second of wavelet deconvolution the resulted fused image has better results. 5.2.2 Case of noise only In this section we will examine the behavior of several deconvolution techniques in the case of noise presence only we will use the most common four types of noise. Salt Pepper noise with density 0.06 Gaussian noise with mean of zero and variance of 0.06 Speckle noise with mean of zero and variance of 0.08 and Poisson noise distribution with mean 10. We will adapt these values for the noises in our paper. After the noise is added to the image the noisy image will be denoised using the deconvolution techniques. After that we will combine the image which has the best result of the blind methods with the image obtained from the Weiner filter. The noisy images for our four noise types are illustrated in fig. 2. 18.104.22.168 Salt Pepper noise In this section we will examine the restoration techniques for restoring the bateau.jpg in presence of Salt Pepper noise only with density of 0.06. The resulted images and the resulted metrics are stated in Tables 3 and 4 below. slide 8: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 232 TABLE 3 RESULTS FOR DECONVOLUTION TECHNIQUES FOR SALT PEPPER NOISE WITH DENSITY 0.06. Filter Mean 33 Median 33 Weiner Bilateral RMSE 19.49 11.46 20.86 18.54 PSNR 22.33 26.21 21.74 22.76 IEF 3.56 10 3.04 3.86 time/m-sec 17 26 12 2085 Restored image TABLE 4 RESULTS FOR DECONVOLUTION TECHNIQUES FOR SALT PEPPER NOISE WITH DENSITY 0.06. Filter Adaptive wavelet Fusion RMSE 21.72 20 9.01 PSNR 21.38 22.1 27.45 IEF 50.24 0.01 11.12 time/m-sec 2876 180 1080 Restored image In Tables 3 and 4 we showed the resulted images and the obtained metrics after denoising the bateau.jpg image by the chosen deconvolution techniques and the results of the fuse technique. We note that the most effective blind technique to remove the Salt Pepper noise is the Median33 with highest PSNR and lowest RMSE so the image resulted from it will be fused with the image resulted from the Weiner deconvolution. We can notice the improvement in the quality of the resulted image and also the improvement in the metrics measured PSNR27.45 and RMSE9.01. 22.214.171.124 Gaussian noise In this section we will examine the restoration techniques for restoring the bateau.jpg in presence of Gaussian noise only with mean of zero and variance of 0.06. The resulted images and the resulted metrics of the deconvolution methods and fusion techniques are stated in the Tables 5 and 6 below TABLE 5 RESULTS FOR DECONVOLUTION TECHNIQUES FOR GAUSSIAN NOISE WITH MEAN OF ZERO AND VARIANCE OF 0.06. Filter Mean 33 Median 33 Weiner Bilateral RMSE 25.11 27.88 23.65 22.95 PSNR 20.13 19.22 20.65 20.91 IEF 4.49 3.58 5.03 5.32 time/m-sec 37 26 130 2049 Restored image slide 9: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 233 TABLE 6 RESULTS FOR DECONVOLUTION TECHNIQUES FOR GAUSSIAN NOISE WITH MEAN OF ZERO AND VARIANCE OF 0.06. Filter Adaptive Wavelet Fusion RMSE 24.64 24.34 16.53 PSNR 20.29 20.4 23.71 IEF 29.6 0.02 6.53 time/m-sec 2978 194 850 Restored image In Tables 5 and 6 we showed the resulted images and the obtained metrics after denoising the bateau.jpg image by the chosen deconvolution and fusion techniques. We note that the most effective technique to remove the Gaussian noise is the Bilateral filter the image obtained from it is fused with the image resulted from the Weiner deconvolution. We can notice the improvement in the quality of the resulted image and the improvement in the metrics measured. 126.96.36.199 Speckle noise In this section we will examine the restoration techniques for denoising the bateau.jpg in presence of Speckle noise only with zero mean and variance of 0.04. The resulted images and the resulted metrics are stated in Tables7 and 8 below. TABLE 7 RESULTS FOR DECONVOLUTION AND FUSION TECHNIQUES FOR SPECKLENOISE WITH VARIANCE 0.04. Filter Mean 33 Median 33 Weiner Bilateral RMSE 19.90 23.59 19.17 19.95 PSNR 21.65 20.67 22.47 22.05 IEF 3.33 2.35 3.54 3.32 time/m-sec 17 23 12 1961 Restored image TABLE 8 RESULTS FOR DECONVOLUTION TECHNIQUES FOR SPECKLE NOISE WITH VARIANCE 0.04. Filter Adaptive Wavelet Fusion RMSE 19.86 20.26 16.89 PSNR 22.17 21.58 24.21 IEF 60.43 0.01 68.17 time/m-sec 2932 189 1125 Restored image In Tables 7 and 8 we illustrated the resulted images and the obtained metrics after denoising the bateau.jpg image by the chosen deconvolution techniques and the fusion technique. We note that the most effective blind technique to remove the Speckle noise is the Adaptive so the image resulted from it been fused with the image resulted from the Weiner slide 10: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 234 deconvolution. We can notice the improvement in the quality of the resulted image also the improvement in the metrics measured. 188.8.131.52 Poisson noise In this section we will examine the restoration techniques for restoring the bateau.jpg in presence of Poisson noise only with mean of 10. The resulted images and the resulted metrics are stated in the Tables 9 and 10 below. TABLE 9 RESULTS FOR DECONVOLUTION TECHNIQUES FOR POISSON NOISE. Filter Mean 33 Median 33 Weiner Bilateral RMSE 14.6269 11.7831 10.7767 16.8529 PSNR 24.8278 26.7056 27.4811 23.5973 IEF 0.5508 0.8371 1.0105 0.4117 time/m-sec 17 24 17 2069 Restored image TABLE 10 RESULTS FOR DECONVOLUTION TECHNIQUES FOR POISSON NOISE. Filter Adaptive wavelet Fusion RMSE 11.3735 15.8774 8.9562 PSNR 27.0129 24.1152 28.9281 IEF 184.053 0.091 54.2584 time/m-sec 2903 183 985 Restored image In Tables 9 and 10 we illustrated the resulted images and the obtained metrics after denoising the bateau.jpg image by the chosen deconvolution techniques and the fusion technique. We note that the most effective blind technique to remove the Poisson noise is the Adaptive filter so the image resulted from is fused with the image resulted from the Weiner deconvolution and we can see that fusion technique has improved the quality of image and the PSNR results. 5.2.3 Case of blur and noise In this section we will take into consideration the presence of blur plus noise. We test the restoration techniques on the bateau.jpg image with the four types of noise and then we will apply the fusion techniques on a two resultant images in order to improve the quality of the resulted image. 184.108.40.206 Blur with Salt Pepper noise Now we will examine the restoration techniques by adding Salt Pepper noise to the blurred image bateau.jpg and then will see the restored images and the resulted metrics. The Salt Pepper will be added with the density of 0.06. Then we will combine the best resulted image from the blind deconvolution methods with the best of the non-blind convolution methods. In Tables 11 and 12 we will show the resulted images and metrics. slide 11: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 235 TABLE 11 RESULTANT IMAGES FOR THE SEVERAL RESTORATION TECHNIQUES WE USED TO REMOVE THE NOISE OF THE SALT PEPPER NOISE WITH DENSITY OF 0.06 AND RESULTANT FUSED IMAGE Degraded image Adaptive BID Bilateral Lucy-Richardson Mean 33 Median 33 Regularized Wavelet Weiner Fused image In the Table 11 we showed the resulted images obtained from the deconvolution methods and the resultant fused image when applied on the degraded image bateau.jpg with the different densities of Salt Pepper type of noise. We can notice the superiority of the Median 33 filter over the other methods also we can notice the better vision of the fused image. TABLE 12 RMSE PSNR IEF AND TIME EXECUTION VALUES FOR THE RESTORED IMAGE OF BLUR PLUS SALT PEPPER NOISE AND THE FUSION TECHNIQUE. Mean 33 Median 33 Weiner Lucy Richardson Bilateral Adaptive Wavelet Regularized BID Fusion RMSE 24.1545 19.04612 21.26462 40.6325 33.6957 27.906 22.8429 60.60693 66.90217 13.46 PSNR 20.471 22.535 21.578 15.953 17.579 19.217 20.955 12.48 11.622 25.34 IEF 2.8588 4.5617 2.9382 1.025 1.1488 30.4938 159.936 0.4494 0.3909 5.35 Time m- sec 6.916 15.083 12.291 63.775 2319.2 2756.2 189.93 237.92 186.68 1142 Table 12 shows the metrics obtained for the Blur with Salt Pepper noise bateau.jpg image restored by the different restoring methods. With the highest PSNR and RMSE the Median 33 is the best filter in restoring image with Blur and Salt Pepper noise. As seen the value of the PSNR and IEF has increased in the fusion image also the value of RMSE has decreased in the fusion image. 220.127.116.11 Blur with Gaussian noise In this section we will examine our deconvolution methods with blur and Gaussian noise presence. In our paper we will consider that Gaussian noise with mean of zero and the value of variance 0.06. After that we will apply the fusion combination on the best resulted image of the blind convolution methods with the best image resulted from the non-blind convolution methods. Resulted images and obtained metrics shown in Tables13 and 14. slide 12: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 236 TABLE 13 RESULTANT IMAGES FOR THE SEVERAL RESTORATION TECHNIQUES WE USED TO REMOVE THE GAUSSIAN NOISE WITH VARIANCE OF 0.06 AND THE RESULTANT FUSED IMAGE. Degraded image Adaptive BID Bilateral Lucy- Richardson Mean 33 Median 33 Rgularized Wavelet Weiner Fused image In Table13 we showed the resulted images obtained from the deconvolution methods when applied on the degraded image bateau.jpg with Gaussian type of noise. We can notice the bilateral filter has the better result. Now we will show the resulted metrics obtained in this section. Fused image is the resulted combination of the Bilateral and Weiner methods. TABLE 14 RMSE PSNR IEF AND TIME EXECUTION VALUES FOR THE RESTORED IMAGE OF BLUR PLUSGAUSSIAN NOISE AND THE FUSION TECHNIQUE. Mean 33 Median 33 Weiner Lucy Richardson Bilateral Adaptive Wavelet Regularized BID Fusion RMSE 28.9903 31.61048 28.00074 68.51861 25.38685 29.30364 26.4214 46.50054 85.48392 21.11 PSNR 18.886 18.134 19.187 11.415 19.996 18.792 19.691 14.782 9.4931 21.63 IEF 3.8843 3.2601 4.1517 0.608 2.8451 21.588 67.212 1.5123 0.4473 5.07 Time/m-sec 7.515 16.386 13.953 119.45 2386.9 21.588 67.212 1.5123 0.4473 695 In Table 14 the results obtained from the bilateral filter has the highest value in PSNR and IEF on the other hand it has the lowest value of RMSE and we notice the improvements of the fusion technique. 18.104.22.168 Blur with Poisson Noise In this section we will examine our deconvolution methods with blur and Poisson noise presence with mean of 10. After that we will apply the fusion combination on resulted image of the blind convolution methods with the best image resulted from the non-blind convolution methods. slide 13: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 237 TABLE 15 RESULTANT IMAGES FOR THE SEVERAL RESTORATION TECHNIQUES WE USED TO REMOVE THE NOISE OF THE POISSON NOISE AND THE FUSED IMAGE Degraded image Adaptive BID Bilateral Lucy-Richardson Mean 33 Median 33 Rgularized Wavelet Weiner Fused image In the Table 15 we showed the resulted images obtained from the deconvolution methods when applied on the degraded image bateau.jpg poisson type of noise. We can notice the Weiner best result compared to other non-blind deconvolution techniques also the bilateral has best result compared to that of blind deconvolution techniques. Now we will show the resulted metrics obtained in this section. Fused image is the resulted combination of the Bilateral and Weiner methods. TABLE 16 RMSE PSNR IEF AND TIME EXECUTION VALUES FOR THE RESTORED IMAGE OF BLUR PLUS POISSON NOISE AND THE FUSION TECHNIQUE. Filter Mean 33 Median 33 Weiner Lucy Richardson Bilateral Adaptive wavelet Regularized BID Fusion RMSE 20.37 19.59 10.70 17.21 11.66 20.67 19.83 45.81 49.72 9.58 PSNR 21.94 22.28 27.53 23.41 26.793 21.82 22.18 14.91 14.19 28.21 IEF 1.1289 1.2314 1.009 0.466 0.853 55.80 318.9 0.29 0.21 1.53 Time/m-sec 8.763 16.57 19.067 67.20 2399.497 2774.292 193.981 91.569 338.679 1125 In Table 16 the results obtained from the bilateral filter has the highest value in PSNR and IEF compared to that of blind deconvolution techniques on the other hand it has the lowest value of RMSE. On the other hand the Weiner deconvolution has the best metrics compared with other non-blind deconvolution methods. We can notice the improvement of the fusion technique on the resultant metrics. 22.214.171.124 Blur with Speckle noise In this section we will examine our deconvolution methods with blur and Speckle noise presence. In our paper we will consider that Speckle noise with mean of zero and variance of 0.04. After that we will apply the fusion combination on resulted image of the blind convolution methods with the best image resulted from the non-blind convolution methods. slide 14: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 238 TABLE 17 RESULTANT IMAGES FOR THE SEVERAL RESTORATION TECHNIQUES WE USED TO REMOVE THE NOISE OF THE SPECKLE NOISE WITH DENSITY OF 0.04 AND THE RESULTED FUSED IMAGE. Degraded image Adaptive BID Bilateral Lucy-Richardson Mean 33 Median 33 Regularized Wavelet Weiner Fused image In the Table 17 we showed the resulted images obtained from the deconvolution methods when applied on the degraded image bateau.jpg with Blur and Gaussian type of noise. We can notice the Bilateral filter has the better result among the blind deconvolution techniques while the Weiner has the better result among the non-blind deconvolution techniques. The fused image is the resulted combination of the Bilateral and Weiner methods. TABLE 18 RMSE PSNR IEF AND TIME EXECUTION VALUES FOR RESTORED IMAGE OF THE SPECKLE NOISE AND THE FUSED RESULTED METRICS. Mean 33 Median 33 Weiner Lucy Richardson Bilateral Adaptive Wavelet Regularized BID Fusion RMSE 22.8283 24.2051 20.225 32.406 19.958 92.669 22.842 54.255 51.115 17.26 PSNR 20.961 20.453 22.48 17.918 22.633 8.792 21.269 13.96 13.442 23.91 IEF 2.3798 1.852 3.0922 1.031 3.1969 3.2761 253.988 0.414 0.3821 4.31 Time/m-sec 7.13 13.894 20.942 880.54 2413.1 2819.3 179.76 197.71 220.63 1002 In Table 18 the results obtained from the bilateral filter has the highest value in PSNR and IEF compared to that of blind deconvolution techniques on the other hand it has the lowest value of RMSE. On the other hand the Weiner deconvolution has the best metrics compared with other non-blind deconvolution methods. We can notice the best results of the fusion technique. VI. CONCLUSION In our paper we made two kinds of comparison the first is a comparison of several restoration techniques in restoring an image that has been degraded with a blur function and several kind of noise. We used several criteria to evaluate the performance of each restoration technique: PSNR RMSE IEF and the time of execution of each of the restoration techniques which has been illustrated in the previous chapter. The Restoration Techniques overall the Bilateral filter has the best results among the several types of noise but in the presence of the Salt Pepper noise the best result it is that been obtained by the median filter from this point of view we can state that the nature of the noise effect the efficiency of the slide 15: International Journal of Engineering Research Science IJOER ISSN: 2395-6992 Vol-2 Issue-5 May- 2016 Page | 239 deconvolution method. On the other hand the blind image deconvolution BID and the regularized filter has offered the lowest results comparing to the rest restoration techniques. On the other hand the efficiency of the blind image deconvolution and the non-blind deconvolutions is very high at blur only images. The proposed combination of two restored images has generated an image with a better vision and higher values in PSNR and lower values in MSE results. REFERENCES 1 A. Kaur and V. Chopra "A comparative study and analysis of image restoration techniques using different images formats" International Journal for Science and Emerging Technologies with Latest Trends vol. 2 no. 1 pp. 7-14 2012. 2 J. Rani and S. Kaur "Image Restoration Using Various Methods and Performance Using Various Parameters" International Journal of Advanced Research in Computer and Software Engineering vol. 4 no. 1 pp. 778-783 January 2014. 3 S. Das J. Saikia S. Das and N. Goni "A COMPARATIVE STUDY OF DIFFERENT NOISE FILTERING TECHNIQUES IN DIGITAL IMAGES" International Journal of Engineering Research and General Science vol. 3 no. 5 pp. 180-191 2015. 4 M. Aggarwal R. Kaur and B. Kaur "A Review of Denoising Filters in Image RestorationBeant Kaur" International Joural of Current research and Academic Review vol. 2 no. 3 pp. 83-89 2014. 5 R. Kaur and N. Singh "A COMPARISON OF IMAGE DENOISING TECHNIQUES FOR HIGH DENSITY SALT--PEPPER NOISE REMOVAL" International Journal For Technological Research In Engineering vol. 2 no. 11 pp. 2602-2607 2015. 6 B. R. Mohapatra A. Mishra and S. K. Rout "A Comprehensive Review on Image Restoration Techniques" International Journal of Research in Advent Technology vol. 2 no. 3 pp. 101-105 2014. 7 U. Qidwai and C. H. Chen Digital Image Processing An Algorithmic Approach with MATLAB Boca Raton: CRC PressTaylor Francis Group 2009. 8 R. L. White "Image Restoration Using the Damped Richardson-Lucy Method" The Restoration of HST Images and Spectra vol. 2 no. 1 pp. 1-7 1994. 9 M. K. Khan S. Morigi L. Reichel and F. Sgallari "Iterative methods of Richardson-Lucy-type for image deblurring" NUMERICAL MATHEMATICS: Theory Methods and Applications vol. 20 no. 10 pp. 1-15 2012. 10 A. Bovik The Essential Guide to Image Processing London: Elsevier 2009. 11 A. Joshi A. K. Boyat and B. K. Joshi "Impact of Wavelet Transform and Median Filtering on Removal of Salt and Pepper Noise in Digital Images" in Internationai Conference on Issues and Challenges in Intelligent Computing Techniques ICICT Ghaziabad 2014. 12 D. Gupta and S. Choubey "Discrete Wavelet Transform for Image Processing" International Journal of Emerging Technology and Advanced Engineering vol. 4 no. 3 pp. 598-602 2015. 13 P. Hedaoo and S. S. Godbole "Wavelet Thresholding Approach for Image Denoising" International Journal of Network Security Its Applications IJNSA vol. 3 no. 4 pp. 16-21 2011. 14 S. Zhong and V. Cherkassky "Image Denoising using Wavelet Thresholding and Model Selection" in Image Processing 2000 Vancouver BC 2000. 15 S. Deswal S. Gupta and B. Bhushan "A Survey of Various Bilateral Filtering Techniques" International Journal of Signal Processing Image Processing and Pattern Recognition vol. 8 no. 3 pp. 105-120 2015. 16 R. C. Gonzalez and R. E. Woods Digital Image Processing vol. 3 New Jersy: Prentice Hall 2008. 17 A. Gota and Z. J. Min "Analysis and Comparison on Image Restoration Algorithms Using MATLAB" International Journal of Engineering Research Technology IJERT vol. 2 no. 12 pp. 1350-1360 2013. 18 P. Kamboj and V. Rani "BRIEF STUDY OF VARIOUS NOISE MODEL AND FILTERING TECHNIQUES" J ournal of Global Research in Computer Science vol. 4 no. 4 pp. 166-171 2013. 19 A. S. Ufade B. K. Khadse and S. R. Suralkar "Restoration of Blur Image Using wavelet Based Image Fusion" International Journal of Engineering and Advanced Technology IJEAT vol. 2 no. 2 pp. 159-61 2012. 20 R. Singh and N. Gupta "Image Restoration Model with Wavelet Based Fusion" Journal of Information Engineering and Applications vol. 3 no. 6 pp. 21-26 2013. 21 N. Mitianoudis and T. Stathaki "Joint Fusion and Blind Restoration for Multiple Image Scenarios with Missing Data" Oxford University Press pp. 1-14 2007. 22 M. Thaku and S. Datar "Image Restoration Based On Deconvolution by Richardson Lucy Algorithm" International Journal of Engineering Trends and Technology IJETT vol. 4 no. 4 pp. 161-165 2014. 23 A. K. Patel and N. Muchhal "Method for Image Restoration using Wavelet based Image Fusion" Internation Journal of Computer Applications vol. 39 no. 13 2012. 24 S. Kaur "Noise Types and Various Removal Techniques" International Journal of Advanced Research in Electronics and Communication Engineering IJARECE vol. 4 no. 2 pp. 226-230 2015. 25 A. Pandey and D. K. K. Singh "ANALYSIS OF NOISE MODELS IN DIGITAL IMAGE PROCESSING" International Journal of Science Technology Management vol. 4 no. 1 pp. 140-144 2015. 26 A. . K. Boyat and B. . K. Joshi "A REVIEW PAPER: NOISE MODELS IN DIGITAL IMAGE PROCESSING" Signal Image Processing : An International Journal SIPIJ vol. 6 no. 2 pp. 63-75 2015. 27 G. Pajares and J. M. de la Cruz "A wavelet-based image fusion tutorial" Pattern Recognition vol. 37 pp. 1855-1872 2004.