Stereoscopic Image Quality Assessment Based on Wavelet Transform and Structure Characteristics

The stereoscopic image distortion can affect the edge, structure, depth and other information of image. In this paper, we propose a no-reference stereoscopic image quality assessment metric based on the human eyes' comprehension of image's low-level structure. First, the left and right views, cyclopean map and disparity map arc decomposed by the dual-tree complex wavelet transform. Second, the phase amplitude characteristics of the wavelet sub-band of the left and right views, cyclopean image and disparity map arc extracted. Similarly the gradient features of the wavelet sub-band of the left and right views and cyclopean image arc extracted. Finally, these features arc feeded into the support vector regression to train the mapping model for predicting the quality score of tested stereoscopic image. The experimental results on LIVE3 DIQD Phase 1 and LIVE3 DIQD Phase 2 show that the proposed method is highly correlated with the human visual system, achieving excellent prediction performance.