Abstract
3D-HEVC introduces advanced intra-frame depth coding tools to enhance compression efficiency, but these come at a high computational cost. This paper proposes a bi-layer texture discriminant algorithm to accelerate depth intra coding. By utilizing a sum of gradient matrix (SGM), the texture complexity (TC) of both the current coding unit (CU) and its subblocks is calculated simultaneously. This dual-layer analysis allows the encoder to skip redundant Depth Modeling Modes (DMMs) and terminate block splitting early. Experimental results show a 44.8% reduction in depth encoding time with a negligible 0.38% increase in synthetic viewpoint bitrate, outperforming existing state-of-the-art methods.
Introduction: The Computational Burden of 3D Video
The 3D extension of High Efficiency Video Coding (3D-HEVC) is the backbone of modern immersive media. It employs the Multi-view Video plus Depth (MVD) format, where depth maps provide the spatial information necessary for view synthesis.
While depth maps primarily consist of large homogeneous regions and sharp object boundaries, their encoding is computationally intensive. 3D-HEVC uses complex tools like Depth Modeling Modes (DMMs) to preserve edges, but the standard encoder's exhaustive search through all possible block sizes and modes is prohibitively slow for real-time applications.
As illustrated in Figure 3, depth map encoding is the dominant factor in total encoding complexity, typically accounting for over 80% of the processing time.
Proposed Method: Bi-Layer Texture Discriminant
The proposed algorithm introduces a "bi-layer" decision mechanism to prune the search space of the HEVC quad-tree structure.
1. Texture Complexity via SGM
The algorithm employs a Sum of Gradient Matrix (SGM) to evaluate the Texture Complexity (TC) of a block. By calculating gradients in four directions ($0^\circ, 90^\circ, 45^\circ, \text{and } 135^\circ$), it identifies whether a region is smooth or contains significant edges.
2. Simultaneous Bi-Layer Analysis
Unlike traditional fast encoders that evaluate blocks recursively, this method calculates TC for the current CU (parent) and its four potential sub-CUs (children) in a single pass.
As shown in Figure 5, HEVC partitions blocks into a hierarchy. The bi-layer approach uses the relationship between parent and child TC values to make early decisions:
- Early Termination (NCU): If the parent and all sub-blocks are smooth, the encoder stops splitting and skips DMM checks.
- Forced Splitting (SCU): If the parent block is highly complex, it is immediately split into smaller CUs.
- Safety Net: The bi-layer check ensures that small details (sharp edges in a sub-block) are preserved even if the parent block appears relatively smooth.
Experimental Results
The algorithm was evaluated using the HTM 16.0 reference software on a variety of 3D video sequences. Key findings include:
- Efficiency: An average reduction of 44.8% in depth encoding time and 36.8% in total encoding time.
- Quality Retention: The average bitrate increase for synthetic viewpoints was only 0.38%, a level virtually imperceptible to viewers.
- Benchmarking: The method demonstrated superior performance compared to other fast-coding algorithms, maintaining a better balance between speedup and compression loss.
Conclusion
By exploiting the specific characteristics of depth maps through simultaneous bi-layer gradient analysis, this research significantly reduces the complexity of 3D-HEVC. The proposed algorithm enables more efficient 3D video processing, bringing real-time 3D encoding closer to practical consumer implementation. Future research may extend these principles to inter-frame coding to further optimize temporal redundancies.
Summary updated