Colorization is an application-oriented technology, which has a long research history. Since the camera was invented, people have tried to colorize black and white photographs to restore the authenticity of photo or achieve some kind of artistic effect. With the development and deepening of information technology, more and more people use computers to assist in handing all kinds of information. Colorization has also been widely used in varieties of areas, such as digital entertainment, animation production, and medical images processing and so on.In academia, some papers on colorization have been published. According to the type of color constraints, these methods can be broadly divided into two categories:one is local color propagation, the other is color transfer. These two methods have advantages and disadvantages, for example, color propagation method requires users to add color brushes as color constraints, but existed methods are sensitive to the initial brush position, if not appropriate, it will cause color bleeding phenomenon. Color transfer method depends on a suitable reference image, and the key for correct color transfer is the features and feature matching chosen. So far, a convenient and effective method has still not been developed, so colorization is an active and challenging research. The thesis mainly study the feature selection and feature matching method for correct color transfer and color bleeding phenomena for local color propagation.For color transfer method, select the mean, standard deviation of intensity, Gabor texture descriptor and SIFT feature as the matching features, and the Euclidean distance is chosen to measure the similarity of the features. On the other hand, choose the edge-sensitive geodesic editing method for local colorization, and apply it to the recoloring of colored images. Experimental results show the good performance of the methods. And, the methods are further applied to simulate the colors of woodcut prints which have rich colors and complicate textures. |