Rachana Dhannawat et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(2), March - April 2020, 2302 – 2309 2302 ABSTRACT This paper proposes a novel filter, which assigns weight selectively and considers only 4 neighbors for calculation. This filter has reduced complexity approximately by 50 % to that of averaging filter and median filter. The proposed technique has improved PSNR by 13% and SSIM by 40% as compared to noisy images. The execution time is reduced by 70% as compared to averaging technique and 93% as compared to median filter. This filter is used for image deblurring as well and the results are improved in terms of PSNR and SSIM by 11% and 1 % respectively. As this filter has improved results for denoising as well as deblurring, it is called as a dual purpose filter. The filter is tested for both gray and colour images and improves results for both. Key words: Don’t care filter, diamond, denoising, deblurring, plus, 4 neighbours. 1. INTRODUCTION Image degradation is an unavoidable process. The image can get degraded even while capturing the image. Degradation can be due to sensor noise, camera misfocus, object or camera motion, atmospheric conditions, vibrations of atoms in receiver devices etc [1]. Image denoising [2] and deblurring [3] are very important research areas in image processing. Many filters are developed for image denoising [4]. Filters used in the literature are averaging, median [5], high boost, Wiener filter [6], Inverse filter, etc. The best results are obtained by nonlocal means filters (NLM) [7]. Many state of art advanced algorithms such as KSVD [8] K-means clustering with Singular Value Decomposition, clustering based dictionaries with locally learned dictionaries KLLD [9], Clustering based Sparse Representation CSR [10], Local Pixel grouping- Principal Components analysis LPG-PCA [11], Nonlocally Centralized Sparse Representation NCSR [12], Block Matching and 3D filtering BM3D [13] are also developed for image denoising. NCSR [12], Iterative Shrinkage Thresholding algorithm (ISTA) [14], Fast Iterative Shrinkage Thresholding algorithm (FISTA) [15] [16], Two Step Iterative Shrinkage Thresholding algorithm (TwIST) [17], Total Variation TV- based [18] iterative algorithms are used frequently for image deblurring. These all algorithms give good results for image denoising however they are iterative, slow as well as complex as compared to basic filters. This paper proposes a new filter which considers only 4 adjacent neighbors of the central pixel in the calculation and assigns more weight to the central pixel. Remaining 4 neighbors are not part of computation hence obviously the filter is fast and less complex. The main objective of this filter is to denoise and deblur the image with minimum complexity and still maintain the quality of the restored image. A basic filter and its different variations are discussed in this paper. Results are compared in terms of PSNR and SSIM [19] and it is observed that PSNR is improved by 13% as compared to noisy image and SSIM is improved by 40% as compared to noisy image. Complexity is reduced nearby about 50% as compared to any basic filters [20]. This filter gives better results for both denoising as well as deblurring. 2. PROPOSED FILTER In most of the filters, equal importance or weight is given to all the pixels under consideration. For example, Averaging filter [4] takes an average of all pixels in the neighborhood. As a consequence performance of the filter degrades as the redundant pixel values are part of the computation and calculations are also more as all pixels are involved. Basic concept of averaging filter is that as it takes the average of neighbouring pixels so the image gets blurred and the effect of noise is reduced. Problem with this filter is that if any one value is out of the range then average will distribute its effect everywhere in the neighbourhood. In this new filter, this problem is solved by selecting only some important pixels and assigning a higher weight to the central pixel. Pixel values selected are those which majorly affect the central pixel. Rests of the neighboring pixels are ignored. Therefore, the number of calculations obviously decreases. Because of selecting the most important responsible pixel values, accuracy increases. This higher weight is denoted as ‘a’ and is assigned the value as 1 percent of standard deviation of noise. If a=0, the filter is equivalent to an averaging filter with only 5 pixels involved in the calculation. The new filter is as given below. x 1/5 x 1/5 (1+a)/5 1/5 x 1/5 x Figure 1: The new filter with dimension 3*3 A New Faster, Better Pixels Weighted Don’t Care Filter for Image Denoising and Deblurring Rachana Dhannawat 1 , Dr. Archana B. Patankar 2 1 Research Scholar, University of Mumbai, India, rachanadhannawat82@gmail.com 2 Professor, Thadomal Shahani Engineering College, University of Mumbai, India, Archana.patankar@thadomal.org ISSN 2278-3091 Volume 9 No.2, March - April 2020 International Journal of Advanced Trends in Computer Science and Engineering Available Online at http://www.warse.org/IJATCSE/static/pdf/file/ijatcse212922020.pdf https://doi.org/10.30534/ijatcse/2020/212922020