Abstract:  This paper proposes a fast depth map intra algorithm based on sum-of-gradient and segment-wise DC coding (SOG-SDC) for reducing the computational complexity of depth map in 3D-HEVC. When processing depth map intra coding, the exhaustive modes searching, the DMMs (depth modeling modes) selection, and the complex decision in segment-wise DC coding (SDC) are time consuming and not always necessary, this paper proposes a sum-of-gradient (SOG) for prediction unit (PU) to prejudge the texture type (flat/non-flat) of coding unit (CU). According to the texture type, the calculation of DMMs can be selectively skipped and the CU splitting process can be earlier terminated. Moreover, for flat texture, the rate distortion list (RD list) can only consists of smooth prediction mode with SDC other than all modes with SDC. The experimental results show that the proposed algorithm can reduce the encoding time by 26.03% compared with HTM16.0, and the bit rate increases by only 0.14%. The subjective quality of the synthesized viewpoint can also show the effectiveness of the proposed algorithm.

Key words: 3D-HEVC, intra prediction, depth map, depth modeling modes, segment-wise DC coding

摘要： 为了降低3D-HEVC中深度图编码的计算复杂度，提出了结合纹理类型预判的深度图帧内快速编码算法SOG-SDC。针对深度图中穷尽模式搜索开销大，采用DMMs（depth modeling modes）模式几率小，以及分段直流残差编码模式（segment-wise DC coding，SDC）判断复杂的问题，通过计算当前预测单元（prediction unit，PU）的梯度和（sum-of-gradient，SOG）来预判编码单元（coding unit，CU）的类型（平坦/非平坦），并根据CU类型进行两方面的优化：一方面，选择性地跳过DMMs模式的检查，并提前终止CU的分割；另一方面，只对全率失真列表中平坦预测模式进行SDC编码，跳过其他候选模式的SDC编码。实验结果表明，在合成视点主观质量基本不变的情况下，该算法相较于HTM16.0在比特率仅增加0.14%的情况下减少了26.03%的编码时间，有效地降低了3D-HEVC视频编码的计算复杂度。

关键词: 3D-HEVC, 帧内预测, 深度图, 深度编码模式, 分段直流编码