Novel nonlinear hybrid filters for image enhancement

Image noise removal and enhancement are important subjects in image processing. Nonlinear techniques for image enhancement and noise reduction challenge the linear techniques by improving image quality while removing noise. The purpose of this thesis is devoted to systematically unifying theory and techniques for mixed noise removal and image enhancement, and to developing new techniques for removing large amounts of mixed Gaussian and impulsive noise while preserving image details. In this thesis, we introduce three new hybrid filters which combine linear and nonlinear filters to produce new hybrid filters capable of removing large amounts of mixed noise. To efficiently use the ambiguous information in an image, both fuzzy set concepts and fuzzy logic operating rules are utilized in the filter design techniques. The three new filters include the single level trained fuzzy filter (SLTF), the multi-level adaptive fuzzy filter (MLAF), and the decision directed window adaptive hybrid filter (DDWAH).The SLTF filter is designed to remove large amounts of mixed noise by combining an impulse filter with a fuzzy filter. The efficiency of the SLTF filter in removing large amounts of mixed noise while preserving image edges is demonstrated. The MLAF filter is an adaptive SLTF filter which uses the local variance of image gray scales to adapt the weights used in the linear portion of the filter to local image statistics. The MLAF filter provides improved visual performance compared to the SLTF filter. The adaptive DDWAH filter uses local statistics to adapt the window size of the filter to local statistics. This approach prevents distortion of small objects in the image, and removes noise more effectively than non-adaptive filters. The experimental results clearly show the improved noise removal performance and good edge preservation properties. Theoretical analysis verifies the measured results.