Research On Texture And Depth Images Coding Driven By View Synthesis

With the development of information technology,digital multimedia is more and more widely emerging in people’s lives.Image and video undoubtedly occupy the main part of the digital multimedia data.Since the digital representation of raw image and video signals requires a huge storage space,various coding algorithms have been proposed to efficiently compress the signals for the convenience of storage and transmission.However,with the increasing demand for various digital formats and applications,the storage and transmission of digital multimedia signals are still facing enormous challenges.In this thesis,focus on the multiview plus depth format in 3D videos,the compression of texture and depth images is investigated in order to improve the quality of the synthesized virtual view.The main contributions and innovative research results are as follows:? A texture and depth image coding scheme is proposed based on the directional block compressed sensing.Considering the inherent orientation information of the texture or edge among the image,the image block is first scanned along the major orientation and then measured by the block-based compressed sensing.At the decoder side,the restored texture and depth images are firstly inverse scanned and then used to synthesize the virtual view.Experimental results demonstrate that the proposed scheme achieves better quality of synthesized view due to the consideration of the directional information.? An auto-covariance analysis based texture and depth image coding scheme is proposed.Firstly,the texture and depth images are featured using the auto-covariance analysis.Then the image coding scheme based on directional discrete cosine transform is proposed.The view synthesis distortion is considered to help the selection of directional mode for the depth image.Experimental results show that the proposed scheme gets better synthesized view with low bitrates.? A quantized dictionary learning method is proposed.Treated as an image,the dictionary atom is compressed by the image compression techniques.We integrate the compression of the dictionary into the process of dictionary learning.Given a bitrate contraint,the dictionary is optimized by ranking the importance of the dictionary atoms.The proposed scheme is applied to the compression of texture and depth image.Experimental results show that the proposed scheme can achieve various tradeoffs between the bit rate of dictionary and the quality of image restoration.Even at low bitrates,it is still possible to achieve high quality of view synthesis.To further verify the proposed scheme,we apply the compressed dictionary for face recognition.Experimental results also show the effectiveness of the proposed scheme.