Development of multiresolution joint bilateral and decision based filtering algorithms for image denoising and implementation in FPGA

Abstract

Digital image processing techniques have gained much importance in recent years due to its wide range of applications in industry, agriculture, remote sensing and medical field. Lot of researches have been conducted in different areas of image processing like enhancement, restoration, segmentation and compression. Noise removal is one of the challenging task in image processing. Images are often corrupted by noise or even damaged due to the defect in image acquisition and transmission system. The contamination on images not only affects their visual quality but also hinders many further high level computer vision tasks such as image coding, recognition, scene understanding and object tracking. Therefore, as an important image processing operation, either as a stand-alone processing or as a pre-processing, it remains one of the most active topic in image processing and draws much attention from the scholars all over the world. The thesis focuses on presenting novel noise removal algorithms for natural images using multiresolution joint bilateral filtering and hybrid decision based filtering approach. The traditional spatial domain based approaches are not capable of eliminating some noises which is obviously different from one another since they have same global features. Multiresolution analysis enables us to work with pixel and its neighbors, which makes the spatial correlation between the pixels easy to implement denoising algorithms. Wavelet transform provides a precise and unifying framework for the analysis and characterization of image at different resolutions. The components of the wavelet transform analysis are shown to be particularly effective in exploiting periodicities of the image sub-bands, as these sub-bands are expressed in different scales and different orientations. newline