Blind distortion classification using content and perception based features

We propose a novel COntent & Perception based features for DIstortion Classification (COPDIC) that can be used for efficient prediction of different distortions that are present in real world imagery. Unlike existing statistical methods, our approach uses human perception to derive features from local block level characteristics to classify common distortion types in images. Given an image with distortions, this paper presents features and a classification methodology that can be used to accurately predict the distortion type (like JPEG, Blur, JP2K, White Noise). The reported classification accuracies compete well with the state-of-the-art techniques for LIVE IQA, TID & CSIQ databases. The proposed technique has low computational complexity and can be employed for real-time applications.