Efficient information-theoretic-statistical equation for image similarity: Comparison with Statistical Similarity and Information-Theoretic Similarity.

Abstract

Several methods of image similarity were introduced to handle different forms of distortions of the image. Most of these techniques are based on statistics, other techniques are information-theoretical that are more modern and more effective than mathematical.

 in this paper, we present a different than its Common technique, a hybrid face image similarity measure by merging statistic and information theory, which is called in this paper as (ITSM), The proposed method (ITSM) Joins three mathematically balanced equations. The first we have newly created in this paper, entropic equation (EE) based on standard information-theoretic measures like (Shannon entropy, joint entropy). The second also newly created, histogram equation (HE) based on (image histogram and joint histogram). The third one is the standard statistic (SSIM). The (ITSM) was tested in Gaussian noise versus (SSIM) and Histogram-Entropic Scale (ITSSIM), where good results were obtained even under a large range of PSNRs. Simulation tests using the MATLAB images, (AT&T and FEI Brazilian face image) databases indicate that the Suggested hybrid technique outperforms the (SSIM) and (ITSSIM) techniques by being able to detect resemblance in faces under very low PSNR.