Restoration Of Ancient Architecture Images Based On Color Intrinsic Decomposition

Abstract:Due to long history and improper preservation, most of the ancient architecture pictures exist in the form of gray images and suffer from varying degree of scratches along with stains. For the reason that work for restore the color and the damaged part of decorating images is heavy, this thesis present a novel scheme for ancient architecture protection by using digital image processing. The research work is discussed as follows:To solve the problem of most historic architecture pictures exist in the form of gray images, we put forward an effective recolor scheme combined with the existing image colorization algorithm. It has the following content:At first, it takes into account that the inconsistent illumination can cause unnatural, false or missing colors before coloring a target image with reference images under different illumination conditions. The effects of illumination inconsistency are eliminated by recovering intrinsic images from a single image. Then, the advantages of example-based colorization algorithm and scribble-based colorization algorithm are combined to improve the accuracy of repairing the color of historic architecture images. Finally, the method is proposed to transfer selective colors from a set of reference images to a target image as it is difficult to find a satisfactory reference image that owning similar style for target image. The experimental results prove that our algorithm is applicable to colorize grayscale historic architecture pictures.To solve the problem of most historic architecture pictures suffer from varying degree of scratches and stains, we put forward an effective image inpainting algorithm which have the following contributions:At first, it extracts the junior characteristic by mood color and the accurate characteristic by texture information of damaged images to find the most similar style of image as the inpainted reference image in a set of images. Then, a proper repair area in the reference image is acquired by shift match and filled in the damaged region by graph cuts. At last, the novel image colorization algorithm is used to deal with the result of the last step, making the inpainted region in accordance with its surrounding. Experimental results show that our restored results of historic architecture pictures improved greatly in vision quality.