Research On Video Compression Algorithm Based On Visual Perception

With the development of networks and the popularity of information technology, multimedia communication plays an important part in people's social life. Due to the large number of data multimedia and video sequences contain, video signals have to be highly compressed to reduce the redundancy for the demand of the narrow bandwidth. Nowadays, the main concern of most video codecs is how to reduce temporal redundancies, spatial redundancies and coding redundancies. However, human eyes are the terminal receptors for all videos, and there still exist a lot of visual redundancies according to the specialty of vision. Then, how to remove visual redundancies is becoming a big challenge in the field of video coding. More and more people began to focus their eyes on perceptual coding methods, which is now becoming one of the hottest developing directions for video compression. The kernel work of this thesis is to research on vision characteristics, establish perceptual coding models, and use them to accomplish effective coding.The first chapter introduced the main image data compression methods and a variety of redundancies, analyzed the key technologies and characteristics of the most famous video coding standards, like MPEG series, H.26x series and AVS, and explained the main work of this thesis.The second chapter listed some great achievements in visual physiology and visual psychology, including luminance sensitivity, frequency sensitivity, color sensitivity for human eyes, generalized visual research status in image processing field, and presents some visual models in video coding area.HVS luminance and color characteristics are particularly studied. Combined with some principles of visual characteristics, the following chapters proposed two perceptual coding models, one for luminance coding and the other for chrominance coding, and tried to control the errors caused by these two models under the visible threshold in order to improve compression performance of video codec without contaminating the video subjective quality. Experiment results show, the two proposed visual models can help to reduce the encoder output bit rate efficiently while guaranteeing the quality of decoded video.At last, a conclusion of this thesis and the discussion of future research directions are given.