An efficient perceived contrast evaluation model for natural images

Abstract

Perceived image contrast is one of the major factors affecting the image quality on displays. Various methods have been proposed to measure the image contrast. However, image contrasts in most of previous works are focused on B/W and defined on simple patterns such as sinusoidal grating. This paper introduces a perceived contrast evaluation model for natural color images. In pursuit of high accuracy, both global and local contrasts are taken into account. Global contrast indicates difference in the perceived luminance and chroma. Local contrast describes the distinguishable degree in image details. In the proposed method, global contrast is calculated based on the dynamic ranges in lightness and chroma. Local contrast is obtained by gradient computations. Both of the global and local contrasts are merged to achieve the perceived contrast. Two types of performance evaluations are performed. They are cross content and within content evaluations. Results of experiments show that global contrast is more effective in the cross content evaluation where the contrast differences between different natural color images are examined. For both of the cross and within content evaluations, the proposed measure yields high value of correlation coefficient with the subjective scores from human visual tests.