Video communication has received tremendous attention from both academia and industry. The conventional hybrid video coding scheme has demonstrated state-of-the-art coding performance. However, its application in video communication still faces several challenges, such as the transmission error of the wireless networks and the power limitation of the hand-held devices. The purpose of this research is to investigate novel algorithms to address the aforementioned issues in conventional hybrid video coding schemes. This work consists of two major parts as follows....[ Read more ]

Video communication has received tremendous attention from both academia and industry. The conventional hybrid video coding scheme has demonstrated state-of-the-art coding performance. However, its application in video communication still faces several challenges, such as the transmission error of the wireless networks and the power limitation of the hand-held devices. The purpose of this research is to investigate novel algorithms to address the aforementioned issues in conventional hybrid video coding schemes. This work consists of two major parts as follows.

The first part investigates the novel applications and further advances of the newly emerged Wyner-Ziv video coding (WZVC) algorithm. Central to the WZVC is the correlation estimation problem. The accuracy of this correlation estimation has big impact on the coding efficiency of WZVC. In this thesis we investigate this problem and propose a universal solution to a general WZVC scheme. Based on progressive learning and convex optimization, the proposed method achieves considerable gain over the existing methods; One promising application of WZVC is low complexity encoding. In this thesis we investigate using Wyner-Ziv successive refinement (WZSR) in such application to improve the coding efficiency. We also conduct rate-distortion analysis and prove that the loss due to decoder motion estimation is limited to less than 2.17dB using the proposed WZSR based scheme. This result, comparing with the existing analysis result, yields that the proposed scheme can gain 4dB in some cases over the original WZVC scheme; WZVC is fiexible in that the decoder can reconstruct the video almost identically even if using different prediction frames. Based on this feature, we propose a novel Wyner-Ziv based bidirectionally decodable (WZ-BID) video coding scheme and the corresponding video streaming system. This system achieves three major functionalities: reverse playback, stream switching and error resilience simultaneously. We also propose an optimal hybrid WZ-BID/H.264 scheme which outperforms the H.264 scheme in all the tested bit rate and packet loss rate.

The second part addresses the error resilience and error concealment problems of video communication. Error resilient transcoding (ERT) is a technique to improve the error robustness of offline compressed video. However, ERT introduces mismatch error and transcoding distortion. In this work we propose a new video slice type called 'L-slice'. Transcoding the traditional P-slices into L-slices improves the error robustness without introducing mismatch error and trancoding distortion. Based on this, we propose a novel offline-transcoding-online-composing scheme which achieves adaptive error robustness to time-varying channel conditions; Error concealment at the video decoder is to recover erroneous picture region based on correctly decoded region in the same frame or the neighboring frames. However, the existing error concealment techniques cannot give satisfactory result when a whole frame is lost. In this thesis, we propose a novel joint forward-backward error concealment method based on optimal linear minimum square error (LMMSE) estimation. Experiments suggest that the proposed method can achieve considerable gain over the state-of-the-art temporal error concealment method in both objective and subjective quality.