On the quality assessment of enhanced images: A database, analysis, and strategies for augmenting existing methods

Abstract

Most methods of image quality assessment (QA) have been designed for QA of degraded images. This paper presents the results of a study designed to investigate whether existing QA methods can be adapted to succeed on enhanced images. We developed a database containing digitally enhanced images and associated subjective quality ratings. Next, we analyzed the efficacy of select QA methods and their reversemode versions in predicting the ratings. Given the fact that an enhanced image makes the original image appear degraded, we tested both normal and reverse-mode versions, where the latter were implemented by specifying the enhanced image as the reference and the original image as the "degraded" image. Our results demonstrate that this reverse-mode approach improves QA of enhanced images. We present a strategy for further improving the QA methods by using measures of contrast, sharpness, and color saturation.

Color-Enriched Gradient Similarity for Retouched Image Quality Evaluation

“…The performance of the proposed method is evaluated in DRIQ database [12]. DRIQ consists of 26 original images and the corresponding 78 retouched images, which are obtained using Photoshop by adjusting contrast, sharpness, brightness, color, or combination of these properties of the original images [12].…”

“…DRIQ consists of 26 original images and the corresponding 78 retouched images, which are obtained using Photoshop by adjusting contrast, sharpness, brightness, color, or combination of these properties of the original images [12]. Difference mean-opinion-score (DMOS) is used as the subjective score.…”

“…The seven traditional metrics include PSNR, SSIM [7], VIF [13], MAD [6], GSM [8], FSIM [14] and GMSD [15], and the two retouched image quality metrics are Refs. [3] and [4]. In order to investigate the individual contributions of structure module and color module in the proposed method, they are further tested separately, which are denoted by GS and COLOR in Table 1.…”

“…Thus far, very few work has been done on the quality evaluation of retouched images [3], [4]. Vu et al [3] proposed to measure contrast, sharpness and saturation changes between the original and retouched images.…”

“…The final quality score of a retouched image was obtained by pooling the three components. In [4], the same authors further improved this model by integrating it into the Most Apparent Distortion (MAD) [6] metric, which is a state-of-the-art image quality metric for degraded images. Although noticeable advances have been achieved by these metrics in evaluating the quality of retouched images, their performances are far from ideal.…”