Study On Image Processing Technique With Fuzzy Arithmetic And Neural Network

With the development of the science and technology and widely usage ofcomputer, technology of digital image processing have come into everyday lifeand have an important effect on everybody. In recent years digital imageprocessing becomes one of hot-topics in the field of scientific researchincreasingly. Technology of image denoising is a significant step in imageprocessing, because the result of the image denoising has impact on theprecision of segmentation. Image segmentation is a basic, most important andpivotal technology in object recognition, image understanding and computervision. The result of segmentation is impact on recognition and understanding.However, these problems have not yet been resolved. Based on literature inexistence, this paper induces fuzzy technology and neural network to imagedenoising and image segmentation, and does some works as follows:(1)Image denoising based on wavelet are studied. This paper uses waveletcoefficient's local variance to replace wavelet coefficient to judge the degree ofinfluence by noise, because wavelet coefficient's local variance denotes thelocal information of wavelet field more actually. It uses Z function to calculatethe fuzzy membership, then uses soft-threshold function to denoise. Thismethod can get the high pulse signal noise rate and good visual result.(2)Adaptive weighting texture segmentation based on separability isstudied. Texture feature are captured by statistical method, then filter thefeature by separability of texture feature. The remainder feature is adaptiveweighted in order to let some feature of big separability to dominant inclustering. Finally FCM method is used to clustering to realize texturesegmentation. Experiments show precise result of segmentation.(3)An Intensified Fuzzy Kohonen Clustering Network(IFKCN) isproposed, for the slow speed of convergence in texture segmentation by FuzzyKohonen Clustering Network(FKCN). It quickens the speed of convergence byadjusting fuzzy membership. Experiments show that IFKCN is faster in speedof convergence and less in alternation number than FKCN in texturesegmentation.(4)Texture segmentation based on Support Vector Machine(SVM) isstudied. SVM has virtues of small sample training and structure riskminimization. Small sample are used in training the SVM, then use trainedSVM to classify the texture feature. Experiments show that using small trainingsample, we can get precise result of texture segmentation.