Image similarity metric (ISIM): a reduced reference image quality assessment approach

Abstract

Reduced reference image quality assessment (IQA) has been highlighted as a reliable methodology for the purpose of quality-of-service (QoS) monitoring in network visual communication applications; as it requires only selective features of the reference image where complete image cannot be accessed (as reference for IQA). The reduced reference IQA problem has been constrained towards extraction of appropriate features (of reference and transformed images) sensitive to various distortion types along with due coherence with human perception. This paper presents an improved reduced reference IQA approach named as 'ISIM: Image Similarity Metric' based on natural image statistics. The proposed IQA approach involves sub-band decomposition of acquired images for features extraction using wavelets. The extracted features are then used to obtain a global distortion measure (Q) as a function of luminance, contrast and divergence parameters. The computational complexity during features extraction is reduced by employing combination of integral image and gradient magnitude approaches. This distortion measure (Q) is then used to formulate ISIM metric as a linear regression with full-reference SSIM. The validation of the proposed metric is carried out by evaluating the image quality for various types of distortions from CSIQ and IVC databases. Simulation results yield a healthy correlation of proposed ISIM metric and the subjective human opinions along with minimal computational complexity in comparison to other state-of-art reduced reference IQA measures.

Efficient image structural similarity quality assessment method using image regularised feature

“…Correspondingly, the no‐reference methods only evaluate the distorted image and do not require any details of original image [2]. The reduced‐reference methods [11, 12] have access only to obtain partial details in the reference image. Consequently, full‐reference methods will be used in this paper.…”