An image response framework for no-reference image quality assessment

This paper proposes an image response framework for no-reference image quality assessment (NR IQA). The framework is an image quality feature extension framework, extending existing image quality features to new NR image quality features. In the framework, a test image is transformed into a number of sub-images through multi-scale preprocessing. The sub-images are taken as image objects, and each object is exerted with multiple local convolution operations as image inputs. Object responses are extracted from the objects under the inputs based on existing image quality features. All the responses compose a global image response feature vector and are mapped to a quality index. Object responses are the NR extensions of the existing image quality features, which signifies the extensions of IQA approaches. Experiments show that the framework can extend full-reference IQA and reduced-reference IQA approaches to NR IQA approaches, and extend NR IQA approaches to new higher-performance NR IQA approaches. (C) 2017 Elsevier Ltd. All rights reserved.