基于分级B预测结构的快速多视点深度编码算法

Fast Multi-View Depth Coding Algorithm Based on Hierarchical B Prediction Structure

多视点深度(MD)通常采用分级B预测结构进行编码,但其编码的复杂度很高.为降低MD的编码复杂度,提出了一种快速编码算法.该算法利用分级B预测结构中较高时间层(TL)内的编码帧与参考帧之间的时间间隔短、时域相关性强,以及编码帧中采用大尺寸模式(LSM)编码的宏块(MB)所占比例高的特点,结合深度图像由大面积平坦区域和边缘组成的特性,对处于不同TL内的编码帧分别制定了不同强度的候选模式精简策略,以跳过不必要的小尺寸模式.为进一步降低编码复杂度,文中针对选用LSM编码的MB提出了模式提前截止策略.实验结果表明,文中算法与原始编码算法相比,在保证MD编码率失真性能的前提下,编码速度可以提高89.61%.

Multi-view depth （MD） is usually coded by using the hierarchical B prediction structure,but the corresponding coding complexity is very high.In order to reduce the coding complexity of MD,a fast coding algorithm is proposed.The proposed algorithm takes into consideration the characteristic of coded frames,namely,the coded frames of the higher temporal layer （TL） in the hierarchical B prediction structure are of higher temporal correlation with their reference frames due to the shorter temporal interval between them,which results in a higher percentage of the macro-blocks （MBs） coded by using the large size mode （LSM）.Then,by considering the feature that the depth image consists of large flat areas and edges,the simplification strategies of candidate MB modes with different intensities are proposed for the coded frames at different TLs,so as to skip the unnecessary small-size MB modes.Moreover,an early termination algorithm for those MBs adopting LSM is proposed to further reduce the coding complexity.Experimental results demonstrate that,as compared with the original coding algorithm,the proposed method reduces the coding complexity by 89.61％,and maintains the rate distortion performance.