Research On Fast Template Matching Algorithm For Ultra-large Images And Its Application

Template matching is to find the most similar part of an image with a template image,and it has a wide range of applications in practical industrial scenarios.With the development of hardware,camera resolution has become higher and higher,and computer storage capacity has become stronger and stronger.In industrial applications,a single image can reach 1GB,and traditional template matching algorithms can no longer meet the real-time requirements of industrial applications.Therefore,researching fast template matching algorithms for ultralarge images is of great practical significance.The main work of this thesis is as follows:(1)A similarity-based Adaptive Pruning ZNCC(AP-ZNCC)fast template matching algorithm is proposed.The algorithm is based on ZNCC and reduces a large number of invalid computations by pruning the search region and skipping many regions where no target exists.It also combines various acceleration strategies,such as image pyramid,integral map,CUDA.The template matching time of ultra-large images exceeding 1GB is reduced to a few seconds,which fully meets the real-time requirements of industrial applications.(2)A two-stage YOLO fast template matching algorithm is proposed.Based on YOLO,this algorithm first divides the original image reasonably,and selects all candidate targets in the first stage of the algorithm,and then performs accurate matching on the candidate targets in the second stage.In terms of the dataset,a variety of data augmentation methods are combined to generate rich datasets,which solves the challenge of collecting large amounts of data in practical machine vision applications.The Copy-Paste method is also used to improve the detection performance of small targets.Applying the object detection algorithm to the template matching task provides a new solution for industrial machine vision.(3)Based on the AP-ZNCC and two-stage YOLO template matching algorithms,a fast template matching system for ultra-large images is designed and implemented.The system implements the two algorithms proposed in this thesis,and has a friendly interface and simple operation.Tests show that this system has good performance and stability,and it can handle template matching of ultra-large images in most scenarios.