Blind Image Quality Assessment Based on Active Learning

Image quality assessment (IQA) is a task of measuring the perceptual quality of a given image and it is a hot study direction in the region of video and image processing and blind image quality assessment (BIQA) model is one of the most utilized full models. Among the existing BIQA models, the CORNIA model has achieved good performance. In the CORNIA model, Ye et. al. utilized d k-means clustering to cluster the pre-processed image patch s, and utilized d the clustering center as a dictionary. However, experimental observations have shown that when the dictionary size is small, the Superiority of CORNIA decreases rapidly. To address this, we introduce active learning strategies into unsupervised character learning, proposing an active character learning (AFL) framework, and apply it to BIQA model.