Fuzzy Logic Based Adaptive Image Denoising

Transcription

1 Fuzzy Logic Based Adaptive Image Denoising Monika Sharma Baba Banda Singh Bhadur Engineering College, Fatehgarh,Punjab (India) SarabjitKaur Sri Sukhmani Institute of Engineering & Technology,Derabassi,Punjab (India) Abstract- In this paper we proposed a new fuzzy logic based approach that detects and removes random valued salt and pepper noise in gray scale digital images. Proposed work is done in two steps, in first step we detect noisy pixels using fuzzy logic, and in second step we replace those noisy pixels using our fuzzy based approach. Our algorithm also includes the histogram based approach for noise removal from the image. We analyze our method for image parameters like PSNR (Peak Signal to Noise Ratio), MSE (Mean Square Error). To illustrate the proposed method, experiments have been performed on standard gray level test image like Lena and results are compared with other popular image denoising methods. The resultsshow that the proposed methodhas relatively good performance with desirable improvement in the PSNR and MSE of the image Keywords-Impulse Noise; Fuzzy Logic; image denoising; PSNR. 1. Introduction Digital gray scale images obtained through various digital products are often corrupted by impulse noise during image acquisition, transmission and reception. Attenuation of noise and preservation of details are two aspects of image processing [2]. Three important types of noise exist: impulse noise, multiplicative noise and additive noise, however, the most annoying noise, which can degrade the image quality is impulse noise. As we know there are many algorithms available for reduction of each type of noise. For example standard median filter (SMF) that is based on order statistic and is nonlinear filter [1] removes the impulse noise but it mistakenly destroys the edges as it is accustomed in replacing the pixels which are undisturbed by noise.. Other methods available are: Rank conditioned median filters (RCM) [4] in which pixels in the filtering window are ranked according to their magnitudes, center weight median filter (CWMF) [1] which emphasis or deemphasize specific input samples, Arak Fuzzy median filter (AFMF) [1] in which the computation time is a compromise and (FBMF) Fuzzy based median filter. Most of these algorithms provide suitable and good results at smaller percentage of noise levels and find difficulty with higher level noises. But our proposed fuzzy based approach for impulse noise removal gives better image quality improvement than above mentioned methods. 2. Proposed method Fuzzy logic based adaptive filter A gray scale image is represented by a twodimensional array where a location (i, j) is a position in image and called pixel. Often the gray scale image is stored as an 8-bit integer giving 256 possible different shades of gray going From black to white pixels can have value in [0-255] integer interval, but some pixels in an image have not correct value and they represent noise having values 0 or255, this noise is known as salt and pepper noise. Salt and pepper noise is not a spreaded noise. We can show basic image denoising model for a gray scale image which removes impulse noises such as 255(pepper) and 0(salt). Figure 1: Basic Image Denoising Model In our context noise detection and removal mechanism will work according to the following steps: 1.3X3 filter mask for removing low level noise. 2. 3X2 filter mask for removing noise from corners and edges. 3. 4X4 filter mask for removing high level noise. 4. Histogram scanning method for further improving the image by enhancing PSNR. Fuzzy logic means to predict data using metadata i.e. to predict something using different probabilities. Here we are considering random valued salt and pepper noise, the filter used is mean filter and we are applying local scanning to the tested image. First of 502

2 all to detect the noise present in the image we have to find out the values of pixels and analyze the difference of that particular pixel value with itsneighboring pixels values. Normally the difference is large in case of salt and pepper noise. To remove this noise we are using a 3x3 filter mask which will scan the whole image and replace the value of pixel with the mean of neighboring ones. During the scanning process if the vote for a particular pixel is greater than the threshold value then it will take that pixel as noise. This scanning matrix removes only low level noise from the image. Since 3X3 scanning matrix won t be able to remove the noise from corners and edges of the image as it won t scan first row, first column, last row and last column therefore we have to use a scanning matrix of the order 3X2 for the same which will remove noise from the corners and edges of the image. This 3X2 scanning makes our algorithm different from the already existing methods. After that 4X4 scanning is done for removing high level noises which occurs due to the increase in the percentage of noise. At last histogram scanning is applied on the image to remove the leftover noise, still present in the image after applying first two scanning. The histogram tells the number of pixels of a specific value. The horizontal axis of graph represents the tonal variations while vertical axis represents number of pixels in that particular tone. In histogram scanning method we are assuming that the pixels having low values are noise and by selecting a threshold level we will remove the noisy pixels from the image by calculating the mean and using the mean method of filtering. This assures good PSNR and MSE compared to other methods. 3. Implementation and Results We have used MATLAB 7 for implementation and analysis of the results for standard LENA image. Results show better performance of proposed method over other ones as shown in table 1 and we analyzetwo parameters:- 1. Peak Signal to Noise Ratio (PSNR) 2. Mean Square Error (MSE) PSNR is given by PSNR= 10log MSE db After the implementation of algorithm on the 8 bit gray scale Lena image of size 512x512 results are shown in figure 2 to figure 5.Also the results have been plotted graphically in figure 6. The results on the LENA image are shown in table 1 where the noise ratio for random valued salt and pepper noise ranges from 5% to 40%. The results from the Table 1 are plotted graphically in Figure 4and it is clear from the graph that the used filter with fuzzy based approach outperforms all the other filters in terms of parameters which we are considered for evaluation. 4. Conclusion Hence it is concluded that proposed fuzzy based image denoising algorithm is able to suppressand removes the salt and pepper noise in digital gray scale image while preserving details across a wide range of impulse noise corruption and gives better performance than other methods such as standard Median, CWM, RCM etc.the future scope is to extend this fuzzy algorithm for the coloured images. References [1] Kh. Manglem Singh Fuzzy Rule based Median Filter for Gray-scale images Journal of Information Hiding and Multimedia Signal Processing April 2011 [2] HaixiangXu, XiaoruiYue An Adaptive Fuzzy Switching Filter for Images Corrupted by Impulse 2009 Sixth International Conference on Fuzzy Systems and Knowledge Discovery [3] ZHANG Hong-qiao, MA Xin-jun, WU-Ning A New Filter Algorithm of Image Based on Fuzzy Logical 2011 IEEE [4] Kh. Manglem Singh and Pravin K. Bora Improved rank conditioned filter for removal of impulse noise from the images IEEE 2002 [5] Mahmoud Saeidi, M. HasanMoradi, FatemehSagafi Filtering Image Sequences Corrupted by Mixed Noise using a New Fuzzy Algorithm 2006 IEEE [6] Tonghan Wang, Xingyi Li An Efficient Impulse Noise Reduction Algorithm 2011 IEEE [7] HaixiangXu, XiaoruiYue An Adaptive Fuzzy Switching Filter for Images Corrupted by Impulse Noise 2009 Where MSE is given by MSE = 1 mn m n i=0 j =0 ( I K) 2 pixel Where for MSE I is input Image i.e. noisy image K is output image after applying proposed algorithm. 503

3 Table 1: Comparative PSNR performance of different methods in filtering Lena image corrupted with 5% to 40% random valued impulse noise Noise % SM CWM RCM Neuvo Arak FBMF Before applying our Algorithm After applying our Algorithm PSNR PSNR PSNR PSNR PSNR PSNR PSNR MSE PSNR MSE Figure 2: Original Lena image Figure 3: Lena image corrupted with 30% random valued impulse noise Figure 4: Output of 3x3 & 3x2 Fuzzy logic based scanning matrix 504

4 Figure 5: Output of 4x4 fuzzy logic based scanning matrix filter & its histogram showing the noisy pixels with small values on extreme left and right 505

5 Figure 6: Output of Histogram based Fuzzy logic scanning filter and its corresponding histogram which shows that noisy pixels has been eliminated 506

6 PSNR in db PSNR Vs percentage of impulse noise sm cwm rcm neuvo arak prop our percentage of impulse noise Figure 7: Graphical Plot showing the comparison of different filters with our proposed fuzzy based filter 507