Multiple Description Coding Based On Redundancy Optimization For 3D Image

In recent years,3D movies have made 3D technology appear in industry and academia.The application of 3D technology is wide in various fields such as industry,education,spaceflight and medical treatment,and it is enriching the content of existing media.Three-dimensional technology is the development direction of multimedia in the future,but the stereo technology is required for the numbers of viewpoint,and the amount of data becomes the problem.Therefore,high-efficiency image compression has become a hot research.In this dissertation,we use the combination of the texture image and the depth image to represent the 3D image.Multiple description coding solves the problem of serious quality declining which causes by network delay and packet loss in the traditional channel.This dissertation applies multiple description method to the encoding and decoding of three-dimensional images.The main work includes:(1)In the three-dimensional image technology,the depth image represents the distance between the object and the camera in the scene,combining with the multi-view images can effectively synthesize the images of any virtual positions,it can display the information of the whole scene completely.Some of the high-frequency components of the depth map do not have much effect on the quality of the synthesized image to a certain extent,especially for the multiple description coding method,it will also cause compression rate transmission waste.In this dissertation,we propose a method of multiple description coding based on DCT for depth map redundancy removal.Because of the characteristics of DCT coefficients,a Lagrange optimization algorithm is proposed to determine how much high frequency components in the DCT domain are removed.At the decoder,an adaptive zero-padding scheme is proposed to reconstruct the depth map.After coding and optimization algorithms,the rate-distortion performance of the multiple description coding method can be effectively improved.(2)The traditional multiple description coding method often does not fully consider the characteristics of human visual system,and the perception for images quality depends largely on the human eyes.In this dissertation,a method of multiple description coding based on human visual system for depth map with the information of JND is proposed.Depth JND is introduced into the multiple description coding of the depth map,and optimizing the amount of redundant information.Due to the characteristics of the human visual system we re-adjust the redundant information of depth map.After coding and optimization algorithms,better reconstruction quality can be obtained at the same code rate.(3)In order to apply the proposed schemes to the 3D image coding and decoding,we construct a multiple JND model combining with the depth maps and texture images of the left and right views.The MJND model can be verified that it can have more distortion by evaluating the quality of the synthesized image.At the same time,combining with the characteristics of the texture image and the proposed MJND model,we propose a coding method for the texture image which makes it have better anti-interference in the transmission process.The depth image is encoded and decoded according to the multiple description scheme based on human visual system.The quality of synthetic image is evaluated,and we can obtain the better visual effects and reconstruction quality both subjectively and objectively.