S-GMOF: A gradient-based complexity reduction algorithm for depth-maps intra prediction on 3D-HEVC

This paper proposes a complexity reduction algorithm for the depth maps intra prediction of the emerging 3D High Efficiency Video Coding (3D-HEVC). The proposed algorithm is focused on reducing the complexity of the Depth Modeling Mode (DMM) 1, which divides a block using a single Wedgelet. Motivated by the fact that the DMM 1 applies an exhaustive evaluation over the possible Wedgelets, the designed algorithm, called Strong Gradient-based Mode One Filter (S-GMOF), applies a gradient filter on the borders of the encoded block and selects a single Wedgelet to be evaluated. The motivational analysis of the proposed filter is presented in details, focusing on selecting the Wedgelet with higher probability to be chosen. For better results, our solution has maintained the DMM 1 refinement process. Experimental analysis showed that S-GMOF algorithm is capable to achieve 8.2% complexity reduction on depth maps prediction, when evaluated under Common Test Conditions (CTC), with an acceptable impact on the synthesized views quality.