Image Mosaic Algorithm Based On Feature Points

As an important branch of computer vision, image mosaic can be used for a set of images with overlapping regions to achieve the goal of full view and high resolution image through registration and fusion. In reality, restricted by objective conditions or image acquisition equipment, all targets of the same scene can not be clearly presented in one image, resulting in spending a lot of time in the stages of image analysis and application. Image mosaic algorithm provides an effective method to solve the above problem, so that the image can maintain a high resolution while presenting 360 degree panorama. At present, image mosaic algorithm has been widely used in space exploration, remote sensing, livelihood and many other fields.Depending on the National Science and Technology Support Project, the paper summarizes image mosaic algorithm in its theories and methods, focusing on the researches on the critical links in mosaic based on feature points matching. The main research includes:(1) Aiming at the feature invariants and position deviation caused by a long run time in feature points extraction when adopting difference of Gaussian pyramid method based on SIFT image mosaic algorithm, an image mosaic algorithm based on the FAST feature points extraction was proposed. In this algorithm, the first step is replacing original SIFT algorithm with feature extraction method based on FAST. This algorithm not only accelerates the matching function time, but also improves the accuracy of feature matching. (2) To deal with the inefficient mismatching elimination of traditional optimization algorithm and without self-adaptability in fine matching stage, an image mosaic optimization algorithm, based on the adaptive clustering K-means++, was proposed. This algorithm, using cluster algorithm, can effectively find the internal relations of the data and take serf-adaptability to the original matching, finally get correct matches. The experimental results show that the algorithm can effectively decrease the image mismatching accuracy rate and enhance the self-adaptability of the matching optimization algorithm. (3) For huge brightness differences, obvious seams and pixel easily-leakage, an image fusion method which based on region is proposed. The non overlapping regions were processed by equal interval interpolation algorithm and the overlapping areas by the method of fading in and out fusion algorithm. Based on the region guidance, the new fusion algorithm can effectively get seamless images and improve visual effect for the matching images.Research results can be applied to study on image mosaic algorithm which based on feature points matching and support related topics.