Structural Similarity Metric with Disparity Map for Stereoscopic Image Quality Assessment

• Main Author: Voo, Kenny Hon Bing
• Format: Thesis
• Language: English
• Published: Universiti Malaysia Sarawak (UNIMAS) 2020
• Subjects: T Technology (General); TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://ir.unimas.my/id/eprint/32162/4/password.pdf; http://ir.unimas.my/id/eprint/32162/

Objective quality assessment is proposed with the intention to substitute the subjective evaluation by using a computational method to measure the quality of images. The perceived quality of digital image is prone to be affected by image distortions such as blurriness. Hence, several types of stereoscopic or 3D image quality metrics were proposed in the past decade to evaluate the quality of stereoscopic images. Most of the existing 3D image quality metrics do not include any of the depth element and do not fully utilized both reference and distorted depth map in assessing 3D images. This factor may affect the accuracy in assessing the quality of stereoscopic images, as the crucial element in stereoscopic images is constrained from the quality assessment. In this thesis, two new methods with depth element are proposed which are STEREO-SSIM and BAS3, in order to investigate the influence of depth element in the stereoscopic images quality assessment. STEREO-SSIM and BAS3 are proposed by employing the depth element into 2D-SSIM method. Likewise, the performance of STEREO-SSIM and BAS3 has been tested by using benchmark stereoscopic image databases in order to analyze the performance of both methods. The experiment results show that the proposed STEREO-SSIM can achieve Pearson correlation of more than 0.8526, while BAS3 can achieve more than 0.8937 for all the benchmark datasets. The improvement of the statistical correlation analysis shows that depth element indeed affect the effectiveness in assessing stereoscopic image quality assessment. Also, based on the correlation analysis, the proposed methods are top ranked as compared with other state-of-the art 2D as well as 3D quality metrics. This research proves that the role of depth element is significant in analyzing 3D images.