An Information Theoretic Criterion for Image Quality Assessment Based on Natural Scene Statistics

Measurement of visual quality is crucial for various image and video processing applications. Traditional measurements convert the spatial data into some feature domain, such as the Fourier domain, and detect the similarity, such as mean square distance or Minkowsky distance, between the test data and the reference or perfect data. In this paper we approach image quality assessment by presenting a novel information theoretic criterion based on natural scene statistics. Using Gaussian scale mixture model in an information theoretic framework, we design an algorithm to compute the minimum perceptual information contained in the images and evaluate the image quality in the form of entropy. Finally, our algorithm is validated with a database set containing 982 images.