Research On Image Matching Based Target Detection Technology And Algorithm Design

Image matching technology is based upon statistical decision theory and belongs to the research field of object recognition. This technology is mainly used to analyze and extract the features of a given image, then by using the extracted features to detect the specified object and determine its exact position in the original image. In this process, the corresponding decision classifiers or criterions are used according to the given image and object to be detected. We need to design effective detection algorithms, so that the work about detection, image alignment, matching and labeling can be handled by computer automatically with high performance and precision.First, this paper studied the theory of image matching technology. We present the respective characteristics of feature-based and gray level-based image matching technologies, and then introduce the template-based image matching technology and analyze its advantages, disadvantages and possible measures of improvement. After that, this paper has a close look at two main template-based matching algorithms: single-template matching and multiple-template matching. We introduce the techniques used in these two kinds of algorithms and also give the details about the implementation flow of algorithm design. Finally, this paper introduces a software platform named Target Matching System, which integrate the developed algorithms and can be used in the scene of target detection.In the scene of single target detection, single-template matching algorithm has its advantage of simple implementation and high detection rate. However, this algorithm is not suitable for the scene of multiple-target detection. This paper studied single template-based detection technology and then proposed a solution to enhance the computing performance by using integral image and Fast Fourier Transform. Then we use this improved algorithm in the scene of multiple-target detection and by using the developed post-processing algorithm, the single-template matching algorithm has good performance to detect similar targets in an image.As for the targets with more geometry deformation, single-template matching algorithm does not have an ideal performance in this application. As a result, we should use more than one template to participate in the process of matching so as to improve the detection performance. This paper studied the multiple-template matching algorithm for its use in the scene of multiple target detection. We use the techniques of Principle Component Analysis and Fast Fourier Transform to detect targets with rotated angle and also to compute the exact angle value.We design and implement software platform for target detection. In this platform, the algorithms introduced in this paper will be integrated so that the algorithms can have a better application environment. This paper will give a brief introduction about its overall design and functions.In the end, in order to verify the effectiveness of the algorithms proposed in this paper, we did the experiments to detect different types of targets in many images. The generated result data in the software platform helped us to analyze the performance of the algorithm.