Paper Info
Title
3D hybrid just noticeable distortion modeling for depth image-based rendering
Abstract
The 3D Just Noticeable Distortion (JND) threshold in essence depends on Human Visual Sensitivity (HVS). This paper carves out a Hybrid Just Noticeable Distortion (HJND) model to measure JND threshold in the framework of Depth Image-Based Rendering (DIBR) for 3D video. The critical differences between 2D and 3D visual perception, depth saliency and geometric distortion, are combined into the HJND model because their significant influence on HVS. To save bit, the HJND model is introduced into the Multi-view Video plus Depth (MVD) encoding framework as a residual filter. After the residue is filtered by HJND and the reference model named Joint Just Noticeable Distortion (JJND), bit saving is achieved up to 28.79% and 23.53%, respectively, and the 3D impaired videos filtered by HJND and JJND have the similar subjective quality. The experiments demonstrate that HJND describes HVS for 3D video more accurately than the state-of-the-art methods.
Year
DOI
Venue
2015
10.1007/s11042-014-2176-y
Multimedia Tools and Applications
Keywords
Field
DocType
3D Just Noticeable Distortion (3DJND),Depth Image-Based Rendering (DIBR),Human Visual Sensitivity (HVS),Depth saliency,Geometric distortion
Residual,Computer vision,Pattern recognition,Reference model,Salience (neuroscience),Computer science,Artificial intelligence,Rendering (computer graphics),Image-based modeling and rendering,Distortion,Visual perception,Encoding (memory)
Journal
Volume
Issue
ISSN
74
23
1380-7501
Citations 
PageRank 
References 
0
0.34
20
Authors
4
Name
Order
Citations
PageRank
Rui Zhong152.83
Ruimin Hu2961117.18
Hongqiang Wang331340.65
Shi-Zheng Wang4778.39