Research On Some Key Problems Of Segmented Image Coding

Image compression is an important subject of modern digital image processing, and it is significant for the popular multimedia applications. The second-generation coding technology based on human visual system (HVS) has attracted worldwide attentions in the past decades because of its excellent performance on compression ratio and visual effect. Segmented image coding (SIC) takes object structure information into account, which separates objects from background based on their shape and codes those segments with different methods. It is an image compression method closer to human visual perception.Visual cortical model, a kind of artificial neural networks, comes from the research on Primates’visual properties. Hence, the research on its working mechanism and applications in image processing has become a hotspot.The dissertation mainly focuses on some key issues of SIC such as region segmentation, representation of contours and regions. First, based on the study of visual neural network and its characteristics, a novel spiking cortical model with neighboring smoothing and gradient sharpening is proposed to segment images. Then, to evaluate the performance of different methods, a set of novel improved criteria are introduced. In addition, representation methods for contours and pixels within irregular region are investigated. At last, experimental results prove the effectivity of irregular segmented region coding. The main work and innovative points of this dissertation are as follows:1. SIC is demonstrated, which is expected to be a potential method based on HVS. Compared with classical image coding methods, irregular segmented region coding and its characteristics, scheme and some challenges are discussed.2. Gradient operator and smooth operator are introduced into visual cortical model, and a novel neural network named as gradient-coupled spiking cortical model is also proposed. Then the model is applied into region segmentation. Experimental results show that the method can capture boundaries of regions accurately, and those pixels within region are homogeneous.3. Evaluation of region segmentation algorithm is to assess different segmentation methods and their results based on the demand of image coding. A set of criteria are proposed, including the intra-region uniformity, inter-region gray contrast, effectivity and smoothness of contours. Experimental results show the criteria can give a mirror to the characteristics of segmentation algorithm and are in accordance with subjective perception.4. Based on analysis of contours property of segmented region, the dissertation explores an effective contour representation method using difference chain code and Huffman code. It could reduce code stream for contours effectively. In addition, representation of pixels within irregular region is developed. Compared with discrete cosine transform (DCT) and shape-adaptive DCT (SA-DCT), the method using polynomial fitting and DCT is presented.