Research On The Shape Deformation And Registration For 2D Image

Along with the rapid development of computer science, artificial intelligence and the research on thought science, people have begun to study how to explain images using computer system in order to understand the real world, and this ought to be a system that is similar to human vision which in this case means computer-vision. The deformation and registration of 2D image is currently a frontier-topic in the field of computer vision. We can obtain better and brand-new perception using image editing techniques. These methods can improve our ability of recognizing, understanding and changing the world with computers which are widely used in Application fields with developmental prospects, such as aerospace industries, biomedical engineering, industrial inspection, robot-vision, police-affairs, Justice, military guidance, culture, Arts and so on.The purpose of the deformation and registration of 2D image is minimizing the differences between source image and target image. In this field, a key problem is how to maintain the salient feature of images. In this area, a new method has been designed and adopted for the deformation and registration of images based on adaptive graph nodes, which could not only achieve the effects of current methods, but could also be real-time and efficient. The contributions of this work are:1. An adaptive-graph-nodes-selection-method is developed, where graph nodes are selected unevenly. This new approach can better meet the requirements of image processing, and can solve the problem of it generating unnatural results or detail missing when being applied on defective parts, i.e. slim regions, effectively.2. A new image deformation method is proposed based on the adaptive-graph-nodes-selection-method, which can achieve real-time non-rigid image deformation. Moreover, the runtime-performance of this new method does not become weak, when the effect of deformation has been significantly improved.3. A new registration method of images based on adaptive graph nodes is proposed. One can detect correspondence between images with the method which combines region matching method and SIFT-based matching method, and this registration method works well. It is also able to correct errors incurred by inaccurate correspondences.Experimental results on several synthetic and real-world datasets demonstrate the robustness of the proposed deformation and registration methods, both of which work well with regard to the image processing of computer vision.