Research On Image Registration Based On Image Segmentation And SURF Points

The image registration is an elementary method of the image processing technique,it has been widely applied in all parts of daily life except for military in recent years.The accuracy and extraction speed of selected features is the difficulty of the field in the practice application.It is import to improve the quality of the chosen accuracy and the fusion.SURF is a classic feature extraction algorithm.But some redundant points could be appeared in the character points because of the external disturbance during the space changing in practice,which can decrease the accuracy of matching.To solve this problem,this paper proposes a SURF image registration algorithm based on the super pixel segmentation and information entropy to realize the target of image registration.(1)We studied the relationship between information entropy of image blocks and the number of SURF feature points and introduce the SURF algorithm in details.We involve the uniform image blocks,and calculate the entropy of information in image blocks and the number of SURF characters.By amount of data,it analyses the relevance of SURF characters and entropy of information after departed.It discovers that the character points are more with higher entropy rate image pieces and image texture is sharp relatively.Some character points are few with lower entropy rate,which possibly can be noise points.(2)We propose a SURF image registration algorithm based on the super pixel segmentation and information entropy.Firstly,we divide the image into a number of super pixel blocks,which can be used for calculating the entropy rate.Secondly,the optimal threshold can be received by amount of experiment data.The neat image blocks can be deleted by comparing the volume of the entropy rate in image blocks.Finally,we extract SURF feature points of the remaining image blocks and use the RANSAC algorithm to match these character points.It is tested on the experiments,the result show that the approved algorithm can increase the accuracy and shortens the matching.(3)It designs the image matching system.This system includes the proved algorithm and some basic functions of image matching.