Research On High Precision Image Registration Method Based On Line Segment Feature And ICP Algorithm

Today,with the rapid development of digital image processing technology,the image registration technology,as the premise and basis of image mosaic,image fusion and 3D reconstruction technology,has been a typical problem and technical difficulty in the field of image processing.The image registration have a very wide range of applications in computer vision,remote sensing image processing,medical image analysis and other important fields.However,low-texture image registration is difficult.That’s because its texture features are relatively small and unstable,and often the image contrast is low,and it is very sensitive to noise,which makes the feature detection of low-text rure images is difficult,and finally affects the overall image registration accuracy.The article first reviews the development history and research status of the main image registration methods,then focuses on the principle of feature-based image registration methods and compares the current main feature detection algorithms.Afterwards,aiming at the problems in the registration of low-texture images mentioned above,this research proposes a high-precision image registration algorithm called LRI which based on line segment features and ICP(Iterative Closest Points)algorithm.The algorithm first preprocesses the image,and then uses the EDlines(Edge Drawing Lines)algorithm and the line segment descriptor LBD to perform feature detection and feature description on the image respectively.In the subsequent feature matching and solving in the transformation model step,the algorithm improve the traditional image registration method based on the LBD algorithm and use the K-nearest neighbor matching algorithm to initially match the feature point pairs.Then,when there are wrong matches between feature point pairs,the algorithm uses the RANSAC(Random Sample Consensus)algorithm to effectively eliminate the influence of external points when solving the transformation model.Finally,the ICP algorithm is used to iteratively optimize the feature points and accurate the model parameters of the transform,and further reduce the distance between two images after image registration in terms of subpixel accuracy,which effectively improving the registration accuracy of low-texture images.The experimental results show that compared with the traditional image registration method,the image registration algorithm proposed in this study has improved the registration accuracy of low-texture images,and has good robustness to illumination changes,scale transformations,and rotation transformations.