Objective Identification of Most Distorted Coding Units over UHD HEVC Sequence

Franck Chi (1, 2, 3) , Xavier Ducloux (3) , Gérard Madec (2) , John Puentes (1, 2)

1 Lab-STICC_TB_CID_DECIDE
2 ITI - Département Image et Traitement Information
3 TVN - Thomson Video Networks (Entreprise)

This paper identifies most distorted areas using a full reference quality metric and analyzes its coding decisions over Ultra High Definition H.265/HEVC High Efficiency Video Coding encoded sequences for video coding optimization. The quality metric is based on two major functions. First, a coding-defect perceptibility indicator is processed. Second, a local codingdefect indicator is defined by combining structural similarity SSIM and Gabor coherency. Three UHD sequences have been encoded at four bitrates and each quality has been estimated. The area below the 10th percentile in terms of objective quality is identified and respective coding decisions are analyzed. Results suggest that the majority of most distorted CUs share the same coding decisions. 88.05% of most distorted CUs are 32x32 or 64x64, and Merge or Skip mode CUs. However, sequence analysis reveals that most distorted coding mode distribution is content dependent. Therefore, simple global coding mode restriction might not provide the optimal encoding strategy.

Mots clés: High efficiency video coding; Full reference quality metric; Structural similarity; Coding defect perceptibility indicator; Coding decision

Domaines: Théorie de l'information [cs.IT]; Traitement du signal et de l'image [eess.SP]; Applications [stat.AP]

Franck Chi, Xavier Ducloux, Gérard Madec, John Puentes. Objective Identification of Most Distorted Coding Units over UHD HEVC Sequence. ICCE 2015 : IEEE 5th International Conference on Consumer Electronics, Sep 2015, Berlin, Germany. pp.243 - 247, ⟨10.1109/ICCE-Berlin.2015.7391247⟩. ⟨hal-01281211⟩