Research On Mosaic Method For A Class Of Uneven Illumination Image

In daily life, the view scope of Image acquisition equipment is limited. In order to expand fieldof vision, a group of pictures need be spliced into one image. The Panoramic image mosaic play aimportant role for digital image processing, so it has been a hot spot in recent years. However, thephenomenon of uneven illumination make it a challenge for panoramic image mosaic. It wasdifficult to splice the uneven illumination images with traditional method. As we all know, changesin lighting make little influence for color information. In this paper the relationship between thecolor of image and feature extraction has been researched, and it make it possible to splice theuneven illumination images.In this paper, we have analyzed the phenomenon caused by uneven illumination andestablished a panoramic image system which was base on its color information. The system hasgreatly enhanced the success rate for uneven illumination mosaic image. The following aspects itemhas been mainly discussed :1) Research on information recovery of uneven illumination image. This paper presents a newapproach to convert the traditional RGB color information to a kind of color independentcomponent which is base on the Kubelka-Munk theory, and the it was converted to asingle-channel grayscale image. The single-channel grayscale images not only has its originalinformation, but also has a good robustness when the lighting changes rapidly. Experimental resultsverify the effectiveness of the proposed algorithm.2) Research on image feature extraction for uneven illumination image. In this part, a newimage characteristics description method which is based on Harris algorithm has been proposed.The feature extraction play a important role in image mosaic. In order to expand the scope of thefeature, we established a color-based multi-scale Harris detector which is based on Gaussianpyramid image. Experimental results show that the new detector is more suitable for unevenillumination image.3) Research on registration and consolidation for uneven illumination image. In order to reducethe mismatch rate of interest points, this paper presents a voting-strategy-based approach. Comparewith the traditional way which named Local Area Correlation Coefficient to match the interestpoints, it show a great advantage especially when the image was rotated.