结合深度信息的多视点视频编码快速模式选择算法

Fast mode decision algorithm for multi view video coding combined with depth information

摘要:

多视点视频加深度(MVD,multi-view video plus depth)的编码格式包含多个纹理视频序列及其对应的 深度图,深度图与对应的纹理视频具有相 似的边缘信息。 传统的编码模式不考虑两者的联系,单独编码导致复杂度高、编码时间过长。因此,合理 利用深度图与 纹理图的相似性进行编码,可以有效降低编码复杂度,同时应该确保编码质量不受影响。本 文利用深度图 的这一特点辅助纹理视频的编码,提出一种帧间快速模式选择算法,充分利用深度图与纹理 图之间的相似 性,建立一种联合复杂度模型,根据模型得到每个宏块的复杂度。对于复杂度低的宏块,在 计算率失真代 价之前跳过编码中一些不必要的模式,从而降低编码复杂度。实验表明,本文提出的快速模 式选择算法,在保 证率失真性能基本不变的情况下,能减少60.57%的编码复杂度,并最 高能减少80.64％的编码时间。

Abstract:

Multi-view video plus depth format includes several texture videos and corresponding depth maps. Depth maps play an important role in depth image based rendering (DIBR) and have similar edges with texture videos.Conventional coding methods have high complexity and long coding time be cause of ignoring the similarity between the depth and texture images.At the same time,it should be en sured that the quality of coding can not be declined.This paper proposes a fast mode decision algorithm for inter mode decision utilizing this c haracteristic of depth maps.By establishing a model which is named as joint mode complexity,each macro-block will have its own complexity.If the complexity is smaller than a certain value,this macro-block can adopt fast mode decision by skipping some unnecessary modes.The proposed method determines to skip some unnecessary modes in the macro-blocks which have low complexity before calculating rate-distortion cost in order to reduce the coding complexity.Experimental results demonstrate that the proposed method reduces 60.57% computational comple xity of encoder with almost the same R-D performance.The time of coding is declined to a large degree alon g with the reducing of computational complexity.