Wavelet Video Coding with Ladder StructuresR

Good quality compressed video can be obtained with a subband coding structure using a wavelet-based Digital Filter-Bank (DFB). For an eecient implementation , the lter-bank can be decomposed into a ladder structure. The lowest frequency subband is coded with a modiied H.261 codec, and the higher frequency subbands are coded using either Entropy-Constrained Vector Quantization (ECVQ) or Entropy-Constrained Trellis Coded Quantization (ECTCQ). Further results are presented for both tele-conferencing and moderate-motion scenes at bit rates in the 100 to 200 kb/s, and 400 to 500 kb/s ranges, respectively. The design and initial performance results for a subband video coder were presented in 1]. The MPEG bicycles sequence, containing moderate motion , was coded at bit rates in the 400 kb/s range for 10 frames/sec. Most of the higher frequency subbands (HFS) were coded using ECVQ 2]. For two subbands, the ECVQ coding was replaced with ECTCQ 3], and comparisons were made. The coder was adapted for use on an ATM network and preliminary results were given in 4]. In this paper, further results are given for head-and-shoulders images. Additionally , the eeect that the motion compensation quality has on the reconstructed image sequences will be studied. The outline of the paper is as follows. The rest of this section gives a brief overview of the coder. The subband decomposition and the methods for encoding the lowest frequency subband (LFS) and HFS are described. The following section examines the ladder structure, and gives an example of how to decompose a DFB for this implementation. Section 3 considers changes to the motion compensation algorithm to improve performance. Sections 4 and 5 give simulation Figure 1: Frequency Domain Partition for the Seven-Band Decomposition of the luminance color component. results for the bicycles sequence and for a teleconfer-encing sequence, respectively. The following section gives conclusions and areas for further research. The system transforms RGB component video to the Y U V color space in 4:1:1 format. The luminance component then undergoes a seven subband decomposition , and each chrominance component undergoes a four subband decomposition. For the lumi-nance component, the ltering and decimation are rst done in the horizontal direction and then in the vertical direction to create four subbands. The lowest frequency subband is separably ltered and decimated to create four more subbands, yielding a total of seven. Figure 1 shows the frequency domain partition for this component. The smaller subbands are 80 …