TTL-IQA: Transitive Transfer Learning Based No-Reference Image Quality Assessment

Image quality assessment (IQA) based on deep learning faces the overfitting problem due to limited training samples available in existing IQA databases. Transfer is a plausible solution problem, which shared features derived from large-scale Imagenet source domain could be transferred original recognition task intended task. However, and target as well their corresponding tasks are not directly related. In this paper, we propose new transitive transfer method for no-reference image (TTL-IQA). First, architecture of multi-domain developed auxiliary domain, then domain. Second, constructed by generative adversarial network distortion translation (DT-GAN). Furthermore, TTL semantic (SFTnet) proposed optimize TTL-IQA. Experiments conducted evaluate performance various databases, including LIVE, TID2013, CSIQ, LIVE multiply distorted challenge. The results show that significantly outperforms state-of-the-art methods. addition, our demonstrates strong generalization ability.