Recovering Color and Details of Clipped Image Regions

Abstract

In a picture of a bright scene, camera sensor readings may be saturated to a maximal value. This results in loss of variation and color distortion of bright regions in the image. We present an algorithm that exploits data from uncorrupted channels to recover image details and color information. The correction is based on the Color Lines model for image representation in RGB color space. Given an image we identify regions of color clipping, and expand each of them to a color cluster called a Color Line. Then we restore the clipped pixel values of each Color Line according to its unique properties. Our method is suitable for both raw and processed color images, including those with large clipped regions. We also present results of correcting color clipping in video sequences.