Region-Based Color Correction of Images

This paper proposes a region-based color correction method that corrects color-shifted images based on a reference image with a standard background. When correcting images without foreground objects, i.e., objects not in the reference image, both the reference image and the color-shifted image are segmented into several regions. A color correction process based on principal component analysis (PCA) is performed between the corresponding regions on the reference image and color-shifted image. The key concept of the color correction process is to align the color distribution of regions on the color-shifted image with the corresponding ones on the reference image by an Affine transform. The proposed color correction is also applicable to those images with foreground objects. In the experiments, the effectiveness of the proposed method is verified by testing some simple images and real-world images with and without foreground objects. Particularly, the proposed method effectively improves the skin-color area identification in a face detector, which is an important issue in many computer vision applications