No-reference JPEG image quality assessment based on block detection and texture analysis
Yao Lyu, Yingyun Yang · 2016
A new algorithm of no-reference image quality assessment(QA) based on combination of block detection and texture analysis (BDTA) is introduced in this paper which aims at JPEG damaged image. It takes advantage of wavelet sub bands to divide blocks into two types in view of visual perception. Current QA thought is to build relationship between block and subjective assessment, the defect of which is the neglect that specific performance of block is closely related to its background texture. That is the masking effect on block. To indicate the this effect, this paper reckons several statistics derived by Gray-level Co-occurrence Matrix to represent image texture. BDTA is finished via BP neural network learning and is tested by LIVE IQA database and TID2008 database. Experiments are conducted to verify the performance and comparisons with prevalent algorithms are also carried out, which turns out outperforms today's state-of-the-art image quality metrics.