Study On Colored Image Registration Based On Feature

As one of the important research issues in digital image processing, image registration is widely used in military and civil areas. Image registration is the process of overlaying images (two or more) of the same scene taken at different times, from different viewpoints, and/or by different sensors. It is a significant task in the fields of image mosaic, object recognization and so on.The fundamental theory and arithmetic of the image registration are introduced in this paper on the base of reviewing of image registration approaches in domestic and foreign academe. Furthermore, two classical algorithms based on gray scale images are improved. The methods of colored image registration based on feature are the key points in the article. The tests show that the two algorithms based on color image are efficient. This paper completes the following works.1. The research background of image registration is introduced. The research statues in domestic and foreign academe are summarized. The concepts, principles and common methods of feature-based registration and the main framework and process of image registration are studied.2. Improved the SIFT algorithms based on gray scale images and applied it to color image registration. Colored image registration is realized by building scale invariant and Colored SIFT descriptor using RGB color information of the image. The test of the descriptor and image registration proved that Colored SIFT descriptors succeeded to the SIFT descriptors and had qualities of uniqueness and robustness so as to improve the precision of image registration.3. An extended SURF algorithm which can make use of robust color information is proposed. SURF algorithm can only use gray level information rather than color information. Extended SURF algorithm extended the SURF descriptor which is build by the local color characteristics and realized the color image registration. The test of the descriptor and image registration proved that Extended SURF algorithm had a better performance than SURF on scale transformation, rotation, view change and image blur, especially on illumination variation.