Color Restoration Based On Similar Images

Image color restoration is a processing procedure mainly for adding color to grayscale image or black and white images. The processed image is not only maintaining information of the shape of the original gray image, but also having reasonable color information. The technology of image color restoration increase the diversity and the artistic of image processing, and it has been widely used in image editing, historic preservation, film production, virtual reality and other fields. The current study of image color restoration is mainly divided into two categories:interactive color restoration algorithm and color restoration algorithm based on similar image.First of all, this paper studies the research status of color image restoration both at home and abroad to analyze the advantages and disadvantages of several typical color restoration algorithms. And then, we focus on the restoration algorithm based on template image which is similar with the target image. But the main problems with these algorithms are:First, the parameter of the processing procedure is mostly the empirical value, resulting the applicability of algorithm is insufficient; Second, The problem of selecting the template image. In the existing algorithms, the template image is selected by user, which brought the personal subjective impact and labor participation.For the problems of the traditional image color restoration algorithm, a learning framework which based on classification learning for gray-scale image color restoration is proposed. Users need to provide a reference image which semantic similarity to the target image, and then the features of super-pixel from these images are extracted. Building super-pixel-level matching functions to guide the color restoration process by using these features. We extract the gray level co-occurrence matrix properties of the target image for the image classification. For each class, We study the super-pixel matching function is obtained by different parameters This method can not only improve the accuracy of super-pixel matching, but also perform more universal. We also use the information of super-pixel’s neighborhood space to correct the color matching errors.In order to automatically obtain the template image similarity and reduce artificial cost in the color image restoration process, we also studied the algorithm in image retrieval, analyzed the vocabulary tree image retrieval method according to the gray image to retrieve similar images. Based on the basis of the present methods, we join the gray level co-occurrence matrix properties to classify image in the image database. Then we build vocabulary tree according to the image classification. Improving the image retrieval similarity while reducing the retrieval time at the same time. Finally, we design an automated color restoration with image retrieval system.