| Summary: | 碩士 === 國立臺北科技大學 === 電機工程研究所 === 104 === Stereo image quality assessment is to measure the visual difference between two stereo images. Compared to conventional 2D image quality assessment, stereo image quality assessment not only increases the information of assessment, but also considers the human visual perception. Most works in the literature focus on how to be more effective in comparing two images of local information instead of focusing on the fact that human vision has different sensitivity in regions with different contents and distortion. Therefore, the main objective of our work is to consider binocular visual properties to improve the efficient of stereo image quality assessment.
The proposed work is based on the cyclopean image assessment method, which combines left and right images to a cyclopean image with the help of disparity and Gabor feature, and then evaluates it with Information Content Weighted Structural Similarity (IW-SSIM).
In addition, our work uses Binocular Just Noticeable Difference (BJND) model to consider the distortion that human eyes cannot perceive, to promote the underestimated score. Since occlusion may occur due to different view angles between stereoscopic image pairs, we also propose an approach that considers occlusion into the cyclopean image.
Finally, we propose a Spatial Weighted Dissimilarity (SWD) approach that considers salient object regions into the image quality assessment, and make evaluation results in line with the visual quality perceived by the human eyes.
The experimental results tested on stereo image database show that the proposed assessment method can significantly improve the accuracy and monotonicity of stereo image quality assessment as compared to the other methods. Our overall performance outperforms those of the five commonly used assessment methods.
|
|---|
