Permutation entropy based full-reference image quality qssessment

Baqar, Mohtashim * and Lau, Sian Lun * and Mansoor Ebrahim, (2019) Permutation entropy based full-reference image quality qssessment. In: 2019 IEEE International Conference on Visual Communications and Image Processing (VCIP), 1-4 December 2019, Sydney, Australia. (Unpublished)

Preview

Text Lau Sian Lun Permutation Entropy.pdf - Published Version Available under License Creative Commons Attribution Non-commercial. Download (2MB) | Preview

Abstract

Due to the increasing proliferation of multimedia signals, specifically, image, video and their applications in our daily life, it is indispensable to have methods that can efficiently predict the visual quality of images with high measures of accuracy. Image processing procedures often introduce undesirable distortion in images that require fixing; preferably consistent with a human visual system (HVS). Therefore, an image quality assessment(IQA) framework should be highly accurate as well as computationally efficient; making it viable to be used with different image processing applications, especially, with real-time applications. Motivated by the need of appropriate objective models, we propose a novel objective IQA algorithm, namely, Permutation Entropy Deviation Index (PEDI), based on the working principle of permutation entropy (PE). Permutation entropy helps in detecting and visualizing changes related to structures with the correlation between successive samples instead of considering magnitudes of the signal, and since, perception of an image to the HVS changes more because of structural changes in an image rather than that of visible errors i.e. MSE. Therefore, in this work, we have exploited this property to predict image quality efficiently. Moreover, entropy itself is sensitive to variations, whereas the permutation entropy captures pattern variations in an image. Furthermore, each local patch in the distorted image undergoes a different level of distortion due to structural differences. This motivates us to use permutation entropy to exploit the global variations in the local quality map for image quality assessment. With standard deviation as the pooling strategy, we observed that permutation entropy between reference and distorted images could predict image quality with high measures of accuracy. Experimental results on a subjective database, CSIQ, have shown that the proposed model outperforms most of the existing STOA image quality assessment models and highly correlates with subjective judgements.

Actions (login required)

View Item