Research And Realization Of Image Registration Algorithm Based On Feature Point

Image feature detection and matching is a very important part of image processing,which plays a very important role in 3d reconstruction,medical image processing,image retrieval and other fields.It also has become the hotspot in the field of imagery.In order to obtain better matching effect,with interfered by many factors during the process of matching,it is necessary to improve in these areas such as the image acquisition method,preprocessing,matching accuracy and efficiency.This paper mainly focuses on several aspects aforementioned.To solve the problem of inaccurate feature detection for uneven images,an improved KAZE image registration algorithm based on texture suppression is proposed.In the process of constructing nonlinear scale space by KAZE algorithm,embedding the texture suppression into the nonlinear diffusion equation,which will preserve the correct illumination estimation of the original image.Then adaptive gamma correction is applied to the illumination component,and the combination of the reflection component and the illumination component is aims to restore the details and color of the image,which is the previous step of image registration.The experimental results show that the improved KAZE algorithm can detect the characteristics of uneven illumination images more effectively,and the correct matching rate is higher than the traditional algorithm.In terms of feature point matching,the accuracy and real-time feature of the grid-based motion statistical feature matching algorithm need to be enhanced.An improved image matching algorithm based on this technique is proposed.Firstly,the AKAZE algorithm is used to detect the feature points of the image,which has superior performance in real time and feature detection.Secondly,we apply grid motion statistics algorithm to separate correct matching from error matching.Finally,the PROSAC algorithm is employed to filter the matching point pairs to obtain matching point pairs with higher quality.The experimental results show that the registration performance of the improved algorithm is higher than the original algorithm under the influence of illumination,compression and blur.It can maintain high robustness and matching speed meanwhile,which is more conducive to real-time detection of images.