Non-Predictive Multistage Lattice Vector Quantization Video Coding

Abstract

In recent years, the demand for mobile multimedia applications has increased tremendously. Since the volume of the application data such as video is high and the bandwidth of mobile channels is limited, efficient compression techniques are very much required (Ghanbari, 2003) (Sikora, 2005). This research area has attracted many researchers since the last 40 years, and many related works have been done as reviewed in (Sikora, 2005). Generally, video compression technique aims to reduce both the spatial and temporal redundancy of a video sequence. The motion estimation and compensation is a very efficient technique to exploit the temporal redundancy of the video sequence (Sikora, 2005). Thus, it has been used in video coding standards for application in mobile communications such as in H.263 (H.263, 2000) and H.264 (Ostermann et al., 2004). Although this process offers significant gain in coding efficiency, the encoded bitstream suffers from channel errors during transmission in mobile channels which reduces the reconstructed frame quality at the receiver. Motion JPEG2000 (ISO/IEC, 2002), uses Intra-frame video coding only which eliminates the prediction step uses in motion estimation process in the temporal domain. It offers less design complexity, reduces computational load and increases robustness in wireless environments (Dufaux & Ebrahimi, 2004). In another work done in (Akbari & Soraghan, 2003), a video coding scheme has been developed to omit the prediction step in temporal domain for robust video transmission in noisy mobile environment. In that work, the similar high frequency subbands from each frame within a Group of Frame (GOP) are joined to produce a number of group data. Each of the group data is processed using an Adaptive Joint Subband Vector Quantization (AJSVQ). The Adaptive Vector Quantization (AVQ) technique has been developed based on the work presented in (Voukelatos & Soraghan, 97). In the past years, there have been considerable research efforts in Lattice Vector Quantization (LVQ) for image and video compression schemes (Conway & Sloane, 1988) (Barlaud et al., 94) (Kossentini & Smith, 99) (Sampson et al., 95) (Weiping et al., 97) (Kuo et Al., 2002) (Man et. al., 2002) (Feideropoulou et. al., 2007). The choice for LVQ has been for its property to reduce complexity of a vector quantizer. In video coding, the works have been inclined towards using LVQ with motion estimation and compensation process as explained