Multiple Description Image And Video Coding Research

With the wide use of internet multimedia services, the requirments for real-time image and video transmission on network with high quality of service are booming. However, the unavoidable irresponsibility of internet and wireless transmission channel always degrade the quality of internet multimedia services. Multiple Description Coding (MDC) is well-recognized as an efficient error resilient image and video source coding method, with the aim to ensure the roust transmission of the encoded bit-streams over unreliable networks. In this thesis, the development situation and key technology of MDC are firstly introduced in brief, and the dual-tree discrete wavelet transform (DDWT) is then thoroughly studied. To address the shortage of the existing DDWT-based coding method, a new DDWT-based image coding method is proposed. By extending this new DDWT-based image coding method, a novel DDWT-based multiple description image coding method and a novel DDWT-based multiple description video coding method with redundancy control are further proposed.In the existing DDWT-based image coding method, noise shaping algorithm is utilized to reduce the redundancy and then two trees of DDWT are encoded independently. Hence, there are two disadvantages: 1) The noise shaping algorithm does not consider the property of the encoder so that it can not make the best of its advantage. 2) The two trees of DDWT are encoded independently, which does not exploit their similarity and thus result in a low coding efficiency. To overcome these two deficients, a novel DDWT-based image coding method is proposed in this thesis. In our approach, a quad-tree pruning method is firstly incorporated into the noise shaping, which can choose those coefficients with high influence to reconstructed quality to remove the redundancy while the structure of those coefficients obtained by improved noise shaping method can match to the wavelet coefficient encoder. Then, by making full use of the spatial similarity of the two trees, a new Dual X-tree scheme is employed to jointly code the two trees. As a result, the coding efficiency of the proposed method is significantly improved compared with the existing DDWT-based image coding method.Considering that DDWT produces two wavelet trees that are inherently like two descriptions, a DDWT-based multiple description image coding method, called dual x-tree multiple description image coding, is presented. First, the two wavelet trees, which are produced by performing DDWT on the input image and the proposed improved noise shaping method, are used as the basic information of two descriptions, respectively. Then, for each tree, a side noising shaping method is further exploited to find its optimal matching coefficients in the other tree as the redundancy information, respectively. Finally, the above-mentioned two basic informations and their corresponding redundancy information are grouped as two descriptions, which are coded by the proposed dual x-tree image encoder and then transmitted over their channel, individually.By combing the layer multiple description coding and 3D DDWT-based video coding, a novel DDWT-layer-based multiple description video coding is proposed. Each description consists of basic layer and enhancement layer. Basic layer is obtained from the outcome of the video encoder based on 3D wavelet transform with motion compensation, which is used as the basic information of the two descriptions. The enhancement layers of two descriptions are resulted from the pairwise combinations of four wavelet trees, which are produced by applying 3D DDWT on the error between the basic layer and the original image. The proposed method combines video coding based 3D wavelet with motion compensation and 3D DDWT-based video coding to realize the multiple description video coding. It can achieve higher coding performance since it not only makes full use of the rich directional selectivity of DDWT and the correlation between four wavelet trees, but also eliminate the lack of DDWT in the horizontal and vertical direction.In this thesis, three novel methods are proposed: 1) a new DDWT-based image coding scheme by selecting the coefficient based on noise shaping and quad-tree prune algorithm and jointly coding two trees; 2) a novel DDWT-based multiple description image coding method via center and side noise shaping; 3) a new 3D DDWT-based multiple description video coding method. Exprimental results have demonstrated the effectiveness and accuracy of all the proposed methods. In summary, all the proposed methods can serve as good solutions and have high reference value for realizing the efficient and robust image/video coding.