•  
  •  
 

Turkish Journal of Electrical Engineering and Computer Sciences

Author ORCID Identifier

NACIR OMRAN: 0009-0008-1440-3624

IMEN WERDA: 0000-0001-9631-3252

AMNA MARAOUI: 0000-0002-0448-4378

ROSTOM KACHOURI: 0000-0002-9451-4269

HAMDI BELGACEM: 0000-0002-1506-7975

Abstract

The 3D extension of the high-efficiency video coding standard (3D-HEVC) enables efficient multiview video coding and provides increased coding efficiency by jointly compressing texture and depth maps for advanced 3D displays. However, 3D-HEVC coding efficiency performance, crucial for virtual view synthesis and immersive 3D experiences, comes at the expense of a noticeable computational complexity increase in depth map encoding processes, particularly in the coding tree unit (CTU) partitioning module. This inadequacy hinders its adoption in time-sensitive applications. To speed up the encoding process, two distinct segmentation-based strategies for adaptive CTU partitioning in depth maps are adopted in this study. The purpose of each approach is to predict coding decisions without rate-distortion (RD) cost calculation, thus eliminating the extensive computational time used to check all block decision candidates. The first approach is a heuristic-based method that leverages connected component labeling (CCL) to identify homogeneous depth regions and guide coding unit (CU) splitting decisions. The second one, a machine learning-based method, makes use of fuzzy C-means (FCM) clustering combined with directional gradients for finer partitioning adaptation. For each 64 × 64 CTU, a binary split decision vector consisting of 21 elements is generated as output to completely describe the CTU to be encoded by the 3D-HEVC encoder. Experimental results conducted using HTM16.3 reference software demonstrate that both CCL and FCM methods are capable of accurately characterizing CU homogeneity, helping to bypass time-consuming RD-cost calculations. The CCL-based method reduces encoding time by 60.64% on average, with a 1.17% improvement in compression efficiency. The FCM-based approach achieves even better complexity reduction of 62.03% while preserving 3D-HEVC depth map coding efficiency.

DOI

10.55730/1300-0632.4290

Keywords

3D-HEVC, quadtree coding unit partitioning, complexity reduction, connected components, fuzzy C-means, machine learning

First Page

1069

Last Page

1093

Publisher

The Scientific and Technological Research Council of Türkiye (TÜBİTAK)

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Share

COinS