Automatic Color Cast Detection Algorithm for Surveillance Video Sequence Based on Color Histogram Characteristic

Abstract

Color cast defect detection of video image is challenging and of great importance in the field of video quality evaluation, and one of the difficulties therein is how to distinguish a dominant color (caused by a predominant color, such as an image in which blue sky is a basic tone) from an inherent cast (image presenting a cast due to imaging device failure or taken in a special environment, such as underwater). By analyzing color histogram characteristics of different color spaces, in this paper, we propose a fast two-step automatic cast detection algorithm for an original surveillance video sequence. First a video image is detected in a RGB space; then, according to the color distribution of cast-frame in CIELAB color space, an improved clustering algorithm is employed to distinguish a dominant color from an inherent color cast; finally, the residual frames without clearly concentrated chroma histogram can be reevaluated using cast sensitive region to improve the accuracy and reliability. Experimental results show that the proposed algorithm can better correspond to people’s subjective evaluation and achieve valid color cast quality prediction.