No-Reference Quality Prediction of Distorted/Decompressed Images Using ANFIS

Abstract

Assessing quality of distorted/decompressed images without reference to the original image is difficult because extracted features are not exact and complex relationship exists between image features and its visual quality. The paper aims at assessing the quality of distorted/decompressed images without any reference to the original image by developing a fuzzy inference system (FIS). Five benchmark images are decompressed with varied codebook size and divided into different regions. Several statistical features of these regions and mean opinion score (MOS) based quality of images are applied as input and output, respectively of the FIS rule generation process. The parameters of the FIS are tuned to improve accuracy in quality prediction by implementing an adaptive network based fuzzy inference system (ANFIS). The error between the computed output of the FIS (predicted quality) and the supplied target value (quality obtained under ideal conditions of decompression) is minimized using supervised learning algorithm. Quality of decompressed and various noise incorporated test images are predicted without reference to the original image producing output comparable with other no reference techniques. Results are validated with the objective and subjective image quality measures.