Research On De-noise Algorithms In Digital Image Processing

Digital image processing has become an important method of aquiring complex information in an information age, and it has extended the vision of human beings in some extent. At present, digital image processing has been applied comprehensively in lots of fields and acquired enormous social and economic benefits. Therefore, the relative research and application of the digital image processing makes a significant sense for theory and practice.In the field of digital image processing, image de-noise technology is the bottleneck stage after acquring image. The effect of the processing can affect the performance of the successive stages. Thus, the research on effective de-noise algorithms and the improvement of the performance of de-noise are very important for improving the whole performance of the digital image processing system.Based on analyzing the characteristics of typical de-noise algorithms such as switch median de-noise algorithm, fuzzy weight mean de-noise algorithm and the de-noise algorithm based on information entropy, research is taken according to the present defects. Four new de-noise algorithms are proposed and their effects are assessed both in theory and in experiemental examinations, and the advantages of the algorithms are demonstrated. The major innovative research work is as follows:(1) Based on the literatures, the relative researches are summarized, the characteristics and defects of the existing algorithms are analyzed, and the relative basic theories neessary are introduced.(2) Based on the theoretical framework of the existing switch median de-noise algorithm, an improved switch median de-noise algorithm based on relative threshold is proposed. For the problem of parameter tuning, an automatic parameter tuning approach is proposed, which optimizes the tuning of the parameters in the algorithm. The proposed algorithm can remove salt-pepper noise effectively.(3) According to the fact that the estimation of the weight of fuzzy weight mean algorithm is complicated relatively, by combining the single value fuzzy model and the characteristic of the image de-noise technique, single value fuzzy model based on pixel differences is proposed. This model provides an effective method for the estimation of fuzzy weights. Based on the model, two kinds of de-noise algorithms are proposed:one is the improved fuzzy weight mean algorithm for mixed noise, which takes the modified value of pixel differences as the input of the model, ensures the performance of removing mix noise without adding to the complexity of reasoning. Another is the improved fuzzy weight mean algorithm based on noise detection, which takes the taches of noise check and pixel cutting, assures the choice of the input effective, improves the precision of the estimation of the weight and achieves a good restrain effect on salt-pepper noise and mixed noise.(4) Based on analyzing the characteristics of the image, the method of structure of image feature space is defined. By applying the Parzen window density estimate method to the feature space, the Parzen window density estimate method based on image feature space is proposed. According to the problem of choice of window function, an effective method is proposed, which provides the gist for the choice of the window function. Based on the research mentioned above, by combining information entropy and de-noise technique, an unsupervised adaptive image de-noise algorithm based on Parzen window estimation and information entropy is proposed. The essential characteristics of the algorithm is proved to be equal to an adaptive iterative weight mean de-noise algorithm in theory. The relation among the adaptive smoothing algorithm, bilateral filtering algorithm and the proposed algorithm are demonstrated.(5) The innovative results in the thesis are summarized, and the future research directions and guidelines are proposed.