Abstract
Image repaired is one of the important problems of image processing. In this paper, we propose bad-pixel detection and repair scheme for color image sensors based on a new proximity triangle concept. Unlike previous methods that mostly utilize the spatial correlation as identifying the faulty pixels for repair, we further exploit the chromatic correlations among the three fundamental colors (i.e., RGB). By doing so, we can avoid misclassification of normal pixels into faulty pixels and thereby increasing the image’s quality substantially. Experimental results demonstrated than such a scheme is highly effective. Under low bad-pixel density, the Peak-Signal to Noise Ratio (PSNR) can be boosted from the start-of-the-art 41dB to more than 48dB for benchmark image Lena.