| Multi-focus image refers to a series of images focused on the same scene or object in different depth ranges.It has many applications in machine vision,such as image fusion,shape from focus,scene three-dimensional perception,etc.It can be used in mechanical parts defect detection,tooling wear state monitoring,robot three-dimensional perception and navigation,etc.Registration is an essential preprocessing step before using multi-focus images,and accurate registration is the basis and guarantee for subsequent processing and application of multi-focus images.However,the traditional multi-focus image registration method has low accuracy due to the difficulty in detecting the defocused blurred region features of multi-focus images and the limitations of global registration methods.Therefore,this paper studies the related methods of multi-focus image registration,and proposes the registration methods of conventional multi-focus image and moving multi-focus image respectively,and develops the corresponding image registration software.The main research contents of this paper are as follows:1.Conventional multi-focus image registration.To solve the problem of low accuracy of traditional multi-focus image registration method,a conventional multi-focus image registration method based on B＿SIFT and Delaunay triangulation is proposed.B＿SIFT algorithm based on SIFT and optical flow tracing overcomes the difficulty of feature extraction in defocused blurred areas and greatly increases the number of matching features between images;Delaunay triangulation can divide the image into several local regions;Local registration can be realized by barycentric interpolation after mutual matching among local regions.Experimental results show that the proposed conventional registration method can effectively improve the registration accuracy of multi-focus images,and the experimental results of image fusion further verify the effectiveness of the method.2.Moving multi-focus image registration.When the camera is in motion,there may be large displacement among the multi-focus images,which does not satisfy the small displacement hypothesis of optical flow method.Therefore,the conventional multi-focus image registration method previously proposed is no longer applicable.Aiming at the problems of low registration accuracy of traditional registration methods and inapplicability of new registration methods,a “Coarse—Precise” two-stage motion multi-focus image registration method based on iterative progression is proposed.Firstly,the K＿SURF algorithm and homography transformation model are used to complete the image coarse registration and reduce the large displacement between images;After that,local registration method is used for precise registration.Experimental results show that compared with traditional registration methods,the proposed multi-focus motion image registration method can effectively improve the registration accuracy.3.Development of multi-focus image registration software.In order to reduce the use threshold of the two multi-focus image registration methods,save the learning cost of researchers and improve the convenience of multi-focus image registration,a multi-focus image registration software based on MFC is developed on the basis of this research.At first,the functional requirements of the software are analyzed,then the program design and function realization of the software are carried out,and finally the software development is completed after beautifying the interface.The software test shows that the interface of the developed software is simple and beautiful,and it can easily and efficiently realize the accurate registration of multi-focus images. |
