Efficient highlight removal of metal surfaces

• An efficient method for removing highlights from metal surfaces in an image.
• Subjective and objective evaluations on real video surveillance data.
• Comparisons with other state-of-the-art methods.

This paper proposes a novel, simple but fairly effective algorithm based on polynomial calibration function and inpainting method to address the problem of removing large-scale highlights from metal surfaces in an image. Our algorithm works upon an underlying premise that neighboring pixels in space-time with similar intensities should have similar colors. The entire system mainly consists of four components. First, the candidates of highlight areas are identified based on a modified specular free model, and then these candidates are represented in HSV color space. Next, the color channel V is re-calculated by using a luminance calibration formula. Afterwards, highlight areas in H and S color channels are recovered based on a novel inpainting method. Finally, the restored image is obtained by converting the integration results from HSV back to RGB color space. The experimental results on a number of images captured from rail transportation scenario and comparisons with other state-of-the-art approaches demonstrate the effectiveness of the proposed work.