Study On Color Transferring Algorithm Based On Color-texture Information And Multi-dimensional Neighborhood Features

Color transferring adjusts colors of target images based on color characteristics of source ones. It is an emerging technique in a variety of fields, such as computer vision, virtual reality, and so on, and has also become one of the heated issues in image processing. The main goal of this thesis is to study color transferring techniques for grayscale images. It has two main contributions: on the first hand, we analyze the principles and features of the present color transferring algorithms; on the other hand, we present some new opinions and two novel methods, which increase the accuracy and thus improve the effect. The thesis is organized as follows. First we introduce the background of color transferring for grayscale images. Second we conduct research on the current classical color transferring algorithms. Specifically, we compare and contrast their general frames, and the advantages and disadvantages of their pixel matching approaches. We also discuss that statistics is intensively applied to color transferring. Then we present two novel methods to transfer colors. One gives a brand-new frame, which matches pixels amongst images by leveraging color-texture information during the procedure of color transferring and thus greatly increases the matching accuracy. In most circumstances, this algorithm improves the transferring smoothness, and enhances the truth of virtual color images. It has been published in the Journal of Computer Research and Development. The other first explores the characteristics of pixels and their neighborhood in details. Then it introduces a new multidimensional eigenvector on pixels. Further, combining with the Approximate Nearest Neighborhood algorithm (ANN), we present a new pixel-matching approach, based on which the new color transferring algorithm greatly improves the classical algorithms and has a very good effect on color transferring. To summarize, the two algorithms have an advantage of increasing the accuracy for pixel-matching without affecting the execution efficiency. As a result, the virtual synthesized images after color transferring match well with the true ones. We believe that the two new algorithms can be applied to Virtual Reality and be helpful to propel the application of the color transferring technique in reality and that they have a bright future in many other applications, such as medicine, virtual reality, and aerospace image processing as well.