International Journal of Engineering Trends and Technology (IJETT) – Volume 27 Number 4 - September 2015 Image Restoration and Comparative Analysis Parul Gupta1, Rajesh Mehra2 1 M.E. student Department of Electronics & Comm. Engineering, National Institute of Technical Teachers Training & Research Center/Panjab University, Chandigarh, India 2 Associate Prof. Department of Electronics & Comm. Engineering, National Institute of Technical Teachers Training & Research Center / Panjab University, Chandigarh, India ABSTRACT- Digital images suffer blurring from various known and unknown point spread functions. Image restoration from degraded image is very big problem in image processing. Image restoration is a process by which a noisy or degraded image is converted into nearly an original image. Improvement in the appearance of the image is done by image restoration. In this paper a Memetic Algorithm is used to restore different simulated images. The algorithm runs for each position of whole image. The algorithm converges to a globally optimal restoration. Algorithm is based on extended neighbourhood search, so that local search area optimize to achieve high PSNR in less number of iteration. To narrow the solution space population is initialized by hill climbing method. To evaluate the proposed results a number of experiments have been done and quantitatively different standard parameters are employed that are PSNR, UIQI and Q factor. Keywords: Memetic restoration,fitness function algorithm, image I. INTRODUCTION The imaging systems are not ideal. They are having some imperfections because of non-ideal conditions. The observed image by these imaging systems represents the degraded version of original image. Image degradation at the time of acquisition making an image worthless so, no further analysis can be process on this acquired degraded image. Image can be degraded by blur or noise [1, 2]. An image can be degraded by many factors which can be introduced in the form of motion & Gaussian blur and Gaussian noise. The main problem in image restoration is that the images are ill posed in nature, means a small change in input can lead to a large change in output [3]. A degraded image can lose some very important or useful information and the quality of image is also degraded . Usually blur is having low frequency components and noise is having high frequency component, so making restoration problem more complicated because a single filter cannot eliminate both the factors. A low pass filter will suppress the noise term but in this case the blur will be boost up and in case of high pass filter the opposite phenomena will take place. There is a trade off ISSN: 2231-5381 between image deblurring and noise smoothening [4]. Discrete wavelet transform is using as an standard tool to code the images in image processing. So many restoration methods are used in past decades. Image restoration is always been a active research area for the researchers. However the majority of degraded functions are not linear but for formulation it is assumed that degradation functions are linear and image restoration is assumed to be linear in most of the application. It can be modelled as two dimensional convolution between original image and the point spread function(PSF) [5]. Mathematically the degradation model is modelled as (1) Where g(x, y) denote the degraded image ,f(x, y) denote the original image, h(x, y) denote the PSF ,n(x, y) represents the additive white Gaussian noise assuming zero mean and vaiance σ2 and * denote the convolution in two dimension. The degraded image[6] in matrix form can be written as (2) Original Image f(x, y) + Gaussian Filter Degraded /Blurred Image g(x ,y) Gaussian Noise Fig. 1 Degradation Model II.RESTORATION TECHNIQUES To recover the original image from the degraded image a number of methods are applied in past decades. The http://www.ijettjournal.org Page 195 International Journal of Engineering Trends and Technology (IJETT) – Volume 27 Number 4 - September 2015 Wiener filter is one of the most efficient and basic method to restore the images. However most of the time the reason for the degradation is not known so degradation function cannot be expressed by a linear function. Most of the time the reason for the degradation is not known and one has to guess the initial value , the restoration of image in this case is known as blind image restoration. The degradation function cannot be express in linear function form, blind restoration methods face complexity to solve the problem [7]. Linear degradation function applied to the observed image that is corrupted by non linear function cause serious problems like edges cannot be preserved and quality of image will degrade [8]. To preserve the edges of the image a number of regularization techniques have been employed. Edges of the image are smoothened by regularization and adaptive constrained optimization. After that a PSO based technique is used to preserve the edges of the image [2 3]. Edges are also efficiently improved by Sobel Edge Detection [9]. Most of the techniques to restore the images suffer from large computation time. This computation time is minimized by High speed CT image reconstruction using FPGA [10]. Temporal average filtering is used for efficient noise removal but the drawback is ,it requires large number of time frames [11]. Now a day to restore the degraded image efficiently, restoration problem is treated as optimization problem. Efficient restore result are obtained by minimizing the cost function. So many optimization techniques have been employed for image restoration. Optimization algorithms can be represent in two forms one is deterministic and the second one is probabilistic. The characteristic of the solution and their use in problem, having direct relation in deterministic optimization, while in probabilistic only some of the elements are consider from the search space and further computations are processed on it [12, 13]. Genetic algorithm is one of the early optimization technique based on meta heuristics. Individual selection in initial population and size of population are the main success of Genetic algorithm. Good initialization population gives good result while poor initialization of population gives poor result, resulting the Genetic algorithm to converge prematurely. This premature convergence is one of the inherent characteristic of Genetic algorithm. This characteristic of Genetic algorithm makes it incapable to find the solution of numerous problems [14, 15]. Steps of Genetic Algorithm are given below. 1) Initialization includes random generation of a population of chromosome. ISSN: 2231-5381 2) Selection of individuals are based on their fitness value. Chances of selection are more for more fit individual. 3) Reproduction applies the crossover and mutation on selected individual. Fig. 2 Flow steps of Genetic Algorithm 4) Replacement of the old population Individual by the new one. Fig. 3 flow Chart of Memetic Algorithm The extension of traditionally genetic algorithm is called Memetic Algorithm. To reduce the premature convergence of the likelihood a local search technique is used. The local search can be use at any phase or http://www.ijettjournal.org Page 196 International Journal of Engineering Trends and Technology (IJETT) – Volume 27 Number 4 - September 2015 stage in algorithm to optimize the results. Exploitation and exploration are two significant issues in the search technique. To use available existing knowledge to find out the the better solution is come under exploitation and to investigate unknown and new area in search space is come under exploration. Genetic algorithm is having high Exploration power and local search is having high exploitation power. Balance between exploitation and Exploration is made by Memetic Algorithm[14 15]. Genetic and Memetic algorithms are non traditional optimization method to find robust and efficient solution of the problem. Instead of reaching to the global optimum, the aim of Memetic Algorithm is to find sufficient „good‟ solution according to the characteristics of the problem [14]. vector of fixed length . Each bit ( corresponds to position of a particular gene . The third part, , is a positive real vector of fixed length and corresponds to the particular gene and selection is done on basis of this value. For different system models, the algorithm may be completely different. In response to these problems new image restoration algorithms are generally based on Memetic algorithm. In the proposed method Memetic algorithm does not depend only upon the neighbourhood characteristics like the traditional method for processing but the whole image is considered and the effect of processed image is close to the observation of human visual effects. III. PROPOSED TECHNIQUE Different stages of proposed technique are shown in fig. 4. The proposed technique is performed on different matlab simulated test images. The Memetic Algorithm for parent Selection first make an initial population P from the local search area and then do a number of generations. There are many different ways to generate the individual parent of the initial population. In our case, each individual parent is randomly generated by hill climbing method so that their intensity values falls within local search area limit. After each generation, the best new child population replaces the previous parent population P. The new child population P′ is generated from previous population P in the following way. Best individuals of P are copied to P‟ according to their fitness value. The fitness value measures the best individuals that are persevered along the generations. The remaining of P′ is obtain by crossover and local search. Procedural steps Converting image Fig.4 Design steps of Proposed Algorithm 1) Calculate fitness value of each individual. Fitness value is used to initialize the population with in local search area. is used to denote the fitness value. 2) The new subset strength vector of each individual is updated locally and is as follows into a column matrix , where and denote the gene subset vector and ranking coefficient vector and denote the gene subset strength. Individuals are selected randomly by hill climbing method from ,The first and second part, and are a ISSN: 2231-5381 (3) Where is the 0 to 255 gene subset strength and denote the step size globally with in range selected randomly by extended neighbourhood search http://www.ijettjournal.org Page 197 International Journal of Engineering Trends and Technology (IJETT) – Volume 27 Number 4 - September 2015 for each dimension. For each iteration physical component. will be the 3) The selection of each individual is calculated globally with in limit and is as follows. (4) 4) The process repeated till best several number of iteration. is not achieved in a b 5) The global best value of each individual is calculated by evaluation of elimination process. c d Fig.5 Original Test Images a) Squares, b)Peppers, c) Cameraman ,d) Barbara (5) IV. RESULT ANALYSIS The proposed method is implemented on four simulated images are taken from matlab data set. firstly the different four simulated images are passing through a low pass filter with 5 x 5 Gaussian blur with standard deviation of σ = 1.6 later a Gaussian noise(SNR=20dB) is added in the blurred image. The simulation is done on standard test images of “cameraman”, “barbara”, “squares” and “peppers”. In this section the performance is evaluated by three factors. 1) PSNR 2) Universal image quality index (UIQI) 3) Metric Q The ideal values of PSNR and UIQI are + and 1 respectively The simulations are performed on 256 x 256 images. Fig 5 shows the original test images. fig. 6 shows noisy images corrupted by 5x5 Gaussian blur with standard deviation σ = 1.6 and the random Gaussian noise. The produced results are stored in fig. 7. Fig.6 Noisy test images by Gaussian blur 5x5 and sigma=2.1 and adding Gaussian random noise(SNR=20dB) a) Squares, b) Peppers, c) Cameraman , d) Barbara ISSN: 2231-5381 http://www.ijettjournal.org Page 198 International Journal of Engineering Trends and Technology (IJETT) – Volume 27 Number 4 - September 2015 problems. This paper presented a new blind restoration approach to deblur the remote sensing images. Genetic Algorithm performs good for the exploration of search space but convergence process is slow. Local search techniques are good in exploitation, so the convergence is fast. By hybridization of these two ,the performance can be improve. In this paper the hybridization of extended neighbourhood search is applied of local search of Genetic Algorithm is applied. The proposed method has shown the percentage improvement in PSNR is approximately 37% for all four test images that are used in paper. The proposed restoration method has considerable amount of computational complexity in terms of speed. Processing speed of the proposed method could be improved in future work. REFERENCES Fig.7 Restored test images by proposed method a) squares, b)peppers, c) cameraman ,d) Barbara Fig. 7 shows the restored images and it can be seen that proposed method is able to restore the images with less noise or this method is less sensitive to noise.Table 1. shows comparative and evaluated results. One can easily see that in this proposed method PSNR as well as UIQI index are having higher values in presence of noise. Q index is also having similar analysis pattern as like UIQI indices and PSNR and proposed method validates its good performance. Therefore for the evaluation of the real images also this proposed method could be use. TABLE 1 PERFORMANCE ANALYSIS OF DIFFERENT TEST IMAGES IMAGE BEFORE RESTORATION PSNR UIQI AFTER RESTORATION Q PSNR UIQI Q Square 20.30 0.076 0.032 29.37 0.43 35.12 Pepper 19.70 0.062 0.048 26.64 0.24 13.23 Cameraman 18.38 0.054 0.066 25.51 0.31 33.98 Barbara 18.75 0.067 0.76 25.87 0.41 36.23 V. CONCLUSION The information about the degradation function is not known in most of the cases. Image restoration for these cases required very complex tools to solve these ISSN: 2231-5381 [1] Jianjung Zhang, “An Alternating Minimization Algorithm for Binary Image Restoration”, IEEE Transaction on Image Processing,Vol. 21, No. 2, pp 883- 888 ,February 2012. [2] Ratnakar Dash, Banshidhar Majhi, “Particle swarm optimization based regularization for image restoration”, IEEE Transaction on Nature & Biologically Inspired Computing, Vol.3, pp. 1253 – 1257, 2009 [3] S.Kwang Lee and Yo- Sung Ho, “Edge Preserving Image Restoration using Adaptive Constrained Optimization,” TENCON‟ 98. 1998 IEEE Region 10 International conference on global connectivity in energy, computer, communication and control, Vol.1, pp.70-73, 17-19 December 1998. [4] S. Gendy, G. Kothapalli and A. Bouzerdoum, “A Fast Algorithm For Image Restoration using a Recurrent Neural Network with Bound-Constrained Quadratic Optimization,” Seventh Australian and New Zealand Intelligent Information Systems Conference, pp. 18-21, November 2001. [5] Sugreev Kaur and Rajesh Mehra, “High Speed and Area Efficient 2D DWT Processor Based Image Compression,” Signal And Image Processing: An International Journal, Vol. 1, No. 2, pp. 22- 31, December 2010. [6] Shen, Huanfeng, Lijun Du, Liangpei Zhang, and Wei Gong. “A Blind Restoration Method for Remote Sensing Images,”IEEE Geoscience and Remote Sensing Letters,2012 [7] Na Li and Yuanxiang Li, “Image Restoration using Improved Pariticle Swarm Optimization,” Internatinal Conference on Network Computing and Information Security, pp.394-397, 2011. [8] Swati Sharma, Shipra Sharma and Rajesh Mehra, “Image Restoration using Modified Lucy Richardson Algorithm in the Presence of Gaussian and Motion Blur,” Advance in Electronic and Electric Engineering, Vol. 3, No. 8, pp. 1063-1070, 2013. [9 Rupinder Verma and Rajesh Mehra, “Area Efficient FPGA Implementation of Sobel Edge Detector for Image Processing Applications,” International Journal of Computer Applications, Vol.56, No.16, pp. 7-11, October 2012. [10] Payal Agarwal and Rajesh Mehra, “High Speed CT Image Reconstruction using FPGA,” International Journal of Computer Applications, Vol. 22, No. 4, pp. 7-10, May 2011. [11] Nidhi Rastogi and Rajesh Mehra, “Analysis of Savitzky-Golay Filter for Baseline Wander Cancellation in ECG using Wavelets,” International Journal of Engineering Sciences & Emerging Technologies, Vol. 6, Issue 1, pp. 15-23, August 2013. [12] Antonios Matakos, Sathish Ramani and Jeffrey A. Fessler, “Accelerated Edge –Preserving Image Restoration without Boundary Artifacts,” IEEE Transactions on Image Processing, Vol. 22, No. 5, pp. 2019-2029, May 2013. [13] H. A. Sanusi, A. Zubair and R. O. Oladele, “Comparative Assesment of Genetic and Memetic Algorithms,” Journal of http://www.ijettjournal.org Page 199 International Journal of Engineering Trends and Technology (IJETT) – Volume 27 Number 4 - September 2015 Emerging Trends in Computing and Information Sciences, Vol. 2, No.10, pp. 498-508, October 2011. [14] Poonam Garg, “A Comparision between Memetic Algorithm and Genetic Algorithm for the Cryptanalysis of Simplified Data Encryption Standard Algorithm,” International Journal of Network Security & Its Application, Vol. 1, No. 1, pp 34 -42, April 2009. [15] Rakesh Kumar, Sudhir Narula and Rajesh Kumar, “A Population Initialization Method by Memetic Algorithm,” International Journal of Advance Research in Computer Science and Software Engineering, Vol. 3, Issue 4, pp 519-523, April 2013. [16] B atrice Duvel, Jin Kao-Hao and Jose Crispin Hernandaz Hernandez, “A Memetic Algorithm for Gene Selection and Molecular Classification of Cancer,” 11th Proceedings in Genetic and Evolutionary Computation Conference, pp.201-208, 2009. Authors: Dr. Rajesh Mehra: Dr. Mehra is currently associated with Electronics and Communication Engineering Department of National Institute of Technical Teachers‟ Training & Research, Chandigarh, India since 1996. He has received his Doctor of Philosophy in Engineering and Technology from Panjab University, Chandigarh, India in 2015. Dr. Mehra received his Master of Engineering from Panjab Univeristy, Chandigarh, India in 2008 and Bachelor of Technology from NIT, Jalandhar, India in 1994. Dr. Mehra has 20 years of academic and industry experience. He has more than 320 papers to his credit which are published in refereed International Journals and Conferences. Dr. Mehra has guided 70 ME thesis and he is also guiding 02 independent PhD scholars in his research areas. He has also authored one book on PLC & SCADA. He has developed 06 video films in VLSI area. His research areas are Advanced Digital Signal Processing, VLSI Design, FPGA System Design, Embedded System Design, and Wireless & Mobile Communication. Dr. Mehra is member of IEEE and ISTE. Mrs. Parul Gupta: Mrs. Parul is currently pursuing M.E from National Institute of Technical Teachers Training & Research Chandigarh, India. She has completed B.Tech from I.E.T. M.J.P. Rohilkhand University Bareilly (U.P.). She is having six years of teaching experience in Institute of Technology & Management GIDA, Gorakhpur (U.P.). Mrs. Parul‟s interest areas are Image processing, VLSI, Wireless and Mobile Communication and Digital Electronics. ISSN: 2231-5381 http://www.ijettjournal.org Page 200