Entropy based fuzzy classification of images on quality assessment

Referenced image quality assessment methods require huge memory and time involvement, therefore not suitable to use in real time environment. On the other hand development of an automated system to assessing quality of images without reference to the original image is difficult due to uncertainty in relations between features and quality of images. The paper aims at developing a fuzzy based no-reference image quality assessment system by utilizing human perception and entropy of images. The proposed approach selects important features to reduce complexity of the system and based on entropy of feature vector the images are partitioned into different clusters. To assign soft class labels to different images, continuous weights are estimated using entropy of mean opinion score (MOS) unlike the previous works where crisp weights were used. Finally, fuzzy relational classifier (FRC) has been built using MOS based weight matrix and fuzzy partition matrix to establish correlation between features and class labels. Quality of the distorted/decompressed test images are predicted using the proposed fuzzy system, showing satisfactory results with the existing no-reference techniques.