Research On Image De-nosing Based On TSPL Algorithm

We are living in an information age. People receive a large amount of information indaily life. Images are an important carrier of information and a main way of informationdissemination. Research shows about75percent of information is acquired from outsidethrough people’s visual system. Images are more informative content and more intuitivecompared to other media, such as voice and characters. However, one image is oftencorrupted by noise in the procedures of collection, acquisition or transmission. As a basicstep in the image processing system, image denoising will affect the follow-up steps ofimage processing, such as feature extraction, edge detection, segmentation, patternrecognition. So research on the methods of removing noises holds great significancepractically.We have proposed an algorithm, named Triple-subset Partition Based ImageLayer-presentation (TSPL). TSPL algorithm is a new type of digital image transformmethod, which is truly different from the traditional linear transformation. The key conceptof the algorithm is in that the image grayscale function f(x,y), which is comparativelyirregular, is approximated by a series of high-regular grayscale functions gn(x,y). Ouralgorithm shows the feature of fast convergence, so it is a good approximator. On the basisof TSPL algorithm, we propose a salt and pepper noise removal method, which exploits thenature of salt and pepper noise that the gray value of a normal pixel isn’t much differentamong neighborhood in an image, since there exists strong coherence among neighborpixels, while, on the contrary, the pixels suffered from salt and pepper noise are featured asthe gray values much different from their neighbor pixels. In our proposed algorithm, weadopt the voting strategy to process the three-grayscale images that are produced by TSPLalgorithm. To test the removal results, we adopt the parameterized MSSIM, an objective animage quality evaluation method which is based on structural similarity. Our new approachcan well remain detailed information of original images while removing the noise. Bothsubjective and objective evaluations show that its performance is much better than themedian filter.