Research On Adaptive Image Filtering Technology In Wavelet Domain

As a significant task in image processing, image filtering plays an important role inimproving the image quality, enhancing the understanding of images and the high-level imageprocessing, analysis and interpretation. Traditional algorithms can not effectively compromisebetween noise-removal and detail-preservation, thus the performance is not ideal and satisfying.Wavelet transform is now a powerful tool in image processing and wavelet-based algorithms forimage filtering has both the theoretical and practical significance.The thesis mainly studies the key theory and techniques for image filtering in waveletdomain, and aims at seeking for the calculation strategy of adaptive threshold, finally themultiscale geometric theory for image analysis is brought into image filtering. The main contentsand works are as follows:1. The basic concepts, principles and fast algorithm of wavelet transform are describeddetailedly, the structure characteristic of image wavelet coefficients and its distibutionalrules are summarized and analysed, which provided a guidance for wavelet basedimage filtering techniques;2. The main principles and procedures of wavelet based image filtering are introduced,classical thresholds, thresholding functions and the subjective and objective evaluationmechanisms for image filtering are listed and discussed. Through the experiment, theperformance of different algorithms and wavelet bases are validated and compared, thecommon principles for choosing wavelet bases are finally provided;3. An improved locally adaptive filtering strategy based on Bayes estimation is proposed,which takes the statistical distribution of wavelet coefficients and the relativity amongcoefficients into account, based on redundant wavelet transform, the local adaptivethreshold is then calculated by selecting a proper neighboring window and the Bivariatejoint Shrinkage function is brought into the shrinkage procedure. The subsequentexperiments on different types of noisy images validated the improved strategy;4. Aiming at the shortage of Curvelab filtering algorithm, an improved adaptive shrinkagestrategy in curvelet domain is proposed, which primarily calculates the local energy in aneighboring window and then adaptively determines the shrinkage range of the currentcurvelet coefficient. Experimental result shows that the new method can receive bettereffect for images with abundant edges and detail.