Aghito Shankar Manuel, Forchhammer Søren
Technical University of Denmark, COM DTU Department of Communications, Optics and Materials, DTU, 2800 Kgs. Lyngby, Denmark.
IEEE Trans Image Process. 2007 Sep;16(9):2234-44. doi: 10.1109/tip.2007.903902.
A novel scheme for coding gray-level alpha planes in object-based video is presented. Gray-level alpha planes convey the shape and the transparency information, which are required for smooth composition of video objects. The algorithm proposed is based on the segmentation of the alpha plane in three layers: binary shape layer, opaque layer, and intermediate layer. Thus, the latter two layers replace the single transparency layer of MPEG-4 Part 2. Different encoding schemes are specifically designed for each layer, utilizing cross-layer correlations to reduce the bit rate. First, the binary shape layer is processed by a novel video shape coder. In intra mode, the DSLSC binary image coder presented in [3] is used. This is extended here with an intermode utilizing temporal redundancies in shape image sequences. Then the opaque layer is compressed by a newly designed scheme which models the strong correlation with the binary shape layer by morphological erosion operations. Finally, three solutions are proposed for coding the intermediate layer. The knowledge of the two previously encoded layers is utilized in order to increase compression efficiency. Experimental results are reported demonstrating that the proposed techniques provide substantial bit rate savings coding shape and transparency when compared to the tools adopted in MPEG-4 Part 2.
提出了一种用于基于对象的视频中灰度α平面编码的新方案。灰度α平面传达了视频对象平滑合成所需的形状和透明度信息。所提出的算法基于将α平面分割为三层：二进制形状层、不透明层和中间层。因此，后两层取代了MPEG-4第2部分中的单一透明度层。针对每层专门设计了不同的编码方案，利用跨层相关性来降低比特率。首先，通过一种新颖的视频形状编码器处理二进制形状层。在帧内模式下，使用[3]中提出的DSLSC二进制图像编码器。在此通过利用形状图像序列中的时间冗余的帧间模式对其进行扩展。然后，通过一种新设计的方案对不透明层进行压缩，该方案通过形态侵蚀操作对与二进制形状层的强相关性进行建模。最后，提出了三种对中间层进行编码的解决方案。利用先前编码的两层的知识以提高压缩效率。报告的实验结果表明，与MPEG-4第2部分中采用的工具相比，所提出的技术在编码形状和透明度时可大幅节省比特率。
