Full-Reference Image/Video Quality Assessment Algorithms Based on Contrastive Principal Component Analysis

With the rapid development of communication network, a large number of high resolution and high quality images and videos are transmitted in communication network. At the same time, the explosive growth of image and video content requires efficient management of image and video collection, compression, storage and transmission processes. However, during these processes, images and videos can be distorted, resulting in a decrease in perceived quality. Therefore, it is necessary to put forward an efficient and reliable image/video quality assessment (I/VQA)method to guide the process of image and video processing. This paper presents a series of simple and effective Full-Reference I/VQA algorithms based on contrastive principal component analysis(CPCA). Firstly, the CPCA algorithm is used to extract the Contrastive principal components(CPCs) from the reference image and the distorted image, and the features are calculated. Then, BP neural network is trained to make the features fit the image's mean Opinion Score (MOS) or Difference Mean Opinion Score (DMOS). Finally, it is extended to VQA through different temporal pooling and temporal feature extraction. The proposed algorithms perform well on three image quality assessment datasets and two video quality assessment datasets, and in particular, beats all competitors on MCL-V.