Research On Adaptive Image Mosaic Algorithm For 3D Shape Detection

The image mosaic theory is a research hotspot in the field of image processing and computer vision.The progress of this technology has important practical significance for promoting the development of virtual reality,remote sensing image processing and medical image analysis.Three-dimensional shape detection technology is an important part of modern industrial technology,but its detection range is limited by imaging equipment.How to expand the detection range without losing measurement accuracy is a major problem in this field.Using image mosaic technology,software can be used to automatically match multiple images with overlapping regions,which can synthesize common images into a wide field image without reducing the detection accuracy and affecting the shooting speed,thus expanding the field of view and improving detection accuracy.In this paper,the basic principle of three-dimensional shape detection is studied firstly,and the image acquired by the three-dimensional shape detector is taken as the research object,and the theory and method of feature-based image stitching technology are deeply studied.A series of complete solutions are proposed for the current microscopic image mosaic problem,and a detailed principle analysis is given for the key processing processes such as image enhancement,feature extraction,feature matching,transformation matrix retrieval and image fusion.Finally,the splicing of microscopic images was realized by programming in MATLAB.In the research of image enhancement algorithm,for the problem that the quality of microscopic image is not good enough to extract enough effective feature points,two image enhancement algorithms based on gray histogram and frequency domain are studied.The histogram equalization,double histogram equalization,dynamic histogram equalization and other algorithms and frequency domain filtering enhancement algorithms have been fully studied and experimentally analyzed.Combining the experimental results,the method of double histogram equalization and homomorphic filtering is selected for image enhancement.In the research of feature extraction algorithm,this paper studies the principles of SIFT algorithm,SURF algorithm and two kinds of corner detection algorithms,and designs the program to extract feature points.According to the extraction accuracy of image feature points,the extraction speed of feature points,the anti-interference ability and whether the extracted feature points have rotation and translation invariance,the SIFT feature point extraction algorithm is selected to effectively extract the feature points in the image.In the research of SIFT feature point matching algorithm,this paper establishes a data index(K-D tree)for retrieval.The BBF(Best Bin First)algorithm is used to establish the priority sequence,which improves the search efficiency.The existing SIFT algorithm is improved for the problems found in the experiment,which greatly improves the running speed of the algorithm and reduces the redundancy of the feature points.In the research of image registration and image fusion methods,the RANSAC algorithm is used to obtain the transformation matrix of the image to achieve spatial image registration,which proves that RANSAC can eliminate most mismatched feature point pairs.Image fusion is performed using the best stitching method.Finally,the design experiment is carried out to splicing the image containing only the translation relationship and the image containing the translation and zoom relationship to achieve the desired effect and complete the image mosaic.