Research On RGBD Video Sequence Preprocessing And Entropy Coding

3D video technology brings a lot of convenience to human society,but the disadvantages such as its low-quality depth image and huge video data volume still bring a lot of obstacles to its further development.On the one hand,due to the environment and equipment,the depth image quality of original RGBD video sequences is very poor,which has incomplete information and high noise.On the other hand,people's requirements for video resolution are getting higher and higher,and the amount of data is also increasing.Therefore,it is necessary to preprocess and encode the RGBD video sequence,which is also a key step in the RGBD video system.In preprocessing,based on the correlation of depth image and color image,a local weighted filter is used to fill depth holes and eliminate noise.In video coding,High Efficiency Video Coding(HEVC)standard is established to remove redundant information of video data through various technologies.Entropy coding is one of the key technologies of HEVC,which can effectively remove statistical redundancy of video information.The main research tasks in this paper are the preprocessing of depth image and the entropy coding for video sequences.Based on the correlation of depth image and color image,the geodesic distance is used to measure the similarity of the pixels,and the geodesic filter is used to repair the incomplete depth image.What's more,the transform coefficient entropy coding algorithm is studied.And based on Multiple Sign Bits Hiding(MSBH)algorithm,an improved sign bits hiding algorithm is designed,which reduces the number of coded bits for sign bits and improves the coding performance.The main work of this paper can be summarized as the following three points.Firstly,an improved geodesic distance calculation method based on edge probability map is designed and implemented.Through the edge detection using structured forests,the edge probability map is obtained and the edge probability is used as the inter-pixel weight.Then Dijkstra algorithm is used to calculate the geodesic distance between two pixels,and the caculation method is tested with complexity analysis.Secondly,an improved depth image restoration algorithm using geodesic filter is designed and implemented.The geodesic distance is used as a quantitative index of correlation between pixels and the weight function is designed using Gaussian kernel.So based on this,a local weighted mean filter is designed and further improved.What's more,the proposed algorithm is tested and compared with other algorithms to show its advancement.Finally,an extra sign bit hiding algorithm based on recovery of transform coefficients is designed and implemented.Based on the MSBH algorithm of HEVC,an extra sign bit hiding rule is designed,which not only increase the num of hidden sign bits but also recover the transform coefficient amplitudes changed by MSBH.Because of this,the proposed algorithm reduces both the compression rate and the encoding distortion.The proposed algorithm is implemented on HM-16.0 and tested on a lot of official test sequence,which shows the superiority in terms of encoding performance.In summary,depth image restoration algorithm and entropy coding technology of RGBD video sequences are studied in this paper,and an improved depth image restoration algorithm based on geodesic filtering and an extra sign bit hiding algorithm based on recovery of transform coefficients are proposed.These algorithms improve the quality of RGBD depth images and improve video encoding performance.