Font Size: a A A

Research On Multi-scale Image Matching Algorithm Based On Feature Points And Its Application

Posted on:2018-05-18Degree:MasterType:Thesis
Country:ChinaCandidate:W C FuFull Text:PDF
GTID:2348330542451545Subject:Control engineering
Abstract/Summary:PDF Full Text Request
Image matching is a kind of image processing and image analysis technology,which can be used to determine the corresponding relationship between two images or multiple images with the same scene.It is very important and essential for computer vision and is widely used in the fields of image mosaic,3D reconstruction,object recognition,image retrieval and so on.This thesis concentrates on the multi-scale image matching algorithm that is based on feature points,and the research mainly focuses on the way of detecting and describing of feature points.The main work and innovations can be summarized as follows:1)A novel multi-scale feature point detection algorithm based on fractional-order gradient operator is proposed,in order to preserve more image detail features.Firstly,the nonlinear diffusion based on fractional-order gradient operator is used to build the nonlinear scale space by means of FED algorithm.Then,in the nonlinear scale space,a new feature detector based on Hessian matrix is exploited to compute feature response.Finally,the local maximum points of feature response in the scale space are found as the feature points of the image.The experimental results show that the proposed algorithm can have good performance in geometric and photometric transformations such as image blur,viewpoint change,scale changes,JPEG compression and illumination.2)A novel binary descriptor based on local intensity order(LIOB)is proposed to improve the robustness of binary descriptors to noise and the changes of local structure of an image.Firstly,all the sampling points are selected in the local patch of a feature point.Then,the local intensity order information of each pair of sampling points is compared to generate a bit in the binary descriptor.Finally,the comparison results of all the sampling pairs are grouped together to construct the binary descriptor of this feature point.It is showed that the LIOB descriptor provide better performance than the state-of-the-art binary descriptors.3)The feature point detector and descriptor proposed in this thesis are applied to 3D reconstruction using image sequence.Firstly,the feature point detector and descriptor proposed in this thesis are used to find the correspondence between the sequence images.Then,the method of structure from motion is exploited to restore the camera's motion parameters and get sparse 3D point cloud.Finally,the dense reconstruction of point cloud is realized by using the patched-based multi-view stereo algorithm.The experimental results show that the proposed algorithm is feasible and effective in the application of 3D reconstruction that is based on sequence images.
Keywords/Search Tags:image matching, nonlinear scale space, feature point detection, binary descriptor, 3D reconstruction
PDF Full Text Request
Related items