Automatic Image Stitching Based On Graph Cuts

The image stitching has become a research hotspot in compute vision and image process in recently years. This technique can be widely applied to various fields, such as remote sensing field, medical field, military field and virtual reality technology.The image stitching includes two steps. The first one is image registration and the second one is image compositing. This paper mainly focuses on these two steps and introduces the algorithms and techniques for image registration and image composing.The key step of image stitching is image registration. It is a step in which there is the best alignment of two or more images from different sensors, time, angles or places.There are many ways to achieve image registration, in addition to some of the basic methods, this paper introduces detection, description and matching using image registration techniques based on feature based or especially SURF feature.And we can use the matching characteristics to realize the final image registration changed from transform models.Image fusion is to generate the final stitching images. Some technologies are needed in order to make the final images without stitching seams, blending, ghost and other stitching traces.This paper introduces two ways of fusion based on pixel compositing based and cut seam based and the best dividing line search algorithm based on the dividing line stitching is realized. Graph cuts is the energy function minimization optimization algorithm which is used often in computer vision researches and also people use it in stereo matching, image reconstruction and texture synthesis.In this paper, the graph cuts method is applied to the image stitching used to search for the best image segmentation line, making the image splicing completely without leaving traces.Finally there is a simple image stitching system in this paper and some results and applications of image stitching would be shown too.