Research On Fast Matching Algorithm Based On Image Features And Gray Values

With the universal application of image matching techn ology in the world,image matching algorithm has been paid more and more attention by the global scientific research community.Many experts and scholars have brought the technology related to image matching to study and explore and continue to make new discoveries and breakthroughs.At present,image matching technology has been widely used in many fields,such as remote sensing satellite image recognition,military reconnaissance,medical image diagnosis and verification of face feature information.In or der to optimize the real-time and registration accuracy of image matching algorithm and improve the application efficiency and precision rate of image matching technology,the matching results obtained in this thesis can provide scientific basis for the subsequent work of image analysis by calculating the similarity overlap degree of more than two images under different shooting time,different shooting angle and different shooting environment.Aiming at the shortcomings of image matching,the main research work of this thesis is summarized as the following aspects:(1)When the real time requirement of image matching is high,the matching rate will be reduced.I propose a high speed image matching algorithm based on the trace of Harris autocorrelation matrix and the improved shade value feature.The algorithm uses the trace of the Harris autocorrelation matrix to detect the corner of the unknown image in real time,which can avoid the dependent variables of the Harris responses.Then the image model stored in the computer is used to detect the pixel with the same shade value as the highest order feature point of the sub-region,and calculate the ratio of the grey value of the pixel point and the highest order feature point to its neighborhood gray value.When the ratio is the same,the pixel is defined as the feature point in the image model stored in the computer;finally,the sub-region feature points are matched according to the image model stored in the computer and the unknown image captured in real time.(2)When the unknown image is rotated and blurred,the matching rate decreases.I propose a high-speed image matching algorithm based on the circular window of Harris and Gaussian.The algorithm uses Harris to extract the corner of unknown image.The SIFT feature points are screened in the corner neighborhood,and the low-dimensional descriptors are established by using Gaussian circular window on the selected feature points.This algorithm not only has the high stability of Harris algorithm,but also takes into account the high matching rate of SIFT algorithm for images with large rotation changes.(3)In view of the problem that compression distortion occurs when unknown images are transmitted many times,which leads to the reduction of matching rate,I propose a high-speed image matching algorithm based on improved FAST and FREAK.Firstly,the improved FAST feature point detection template it is constructed for the image model stored in the computer and the unknown image captured in real time.The original image is de-sampled on the basis of the scale space pyramid.In order to avoid the process of convolution between Gaussian kernel and the original image,the improved FREAK algorithm is used to build a dimensional reduce feature descriptor similar to the human retinal structure for each feature point.Finally,the improved Hamming distance is used for feature matching.The algorithm proposed by me has a high matching rate for unknown images with multiple transmission distortion.(4)In order to reduce the matching rate when the scene changes in the neighborhood,I propose a high speed image matching algorithm based on improved Shi-Tomas and regional covariance matrix.The improved Shi-Tomas algorithm is used to screen out the corner points of the unknown imag es,and then the feature model is constructed by using the regional covariance proposed by me.Based on the region covariance matrix,the unknown image captured in real time is matched with the image model stored in the computer.The characteristics of the region covariance matrix ensure the robustness of the unknown image when the Euclidean transformation was applied.Finally,the improved RNASCA algorithm is used to delete the mismatch points,which reduces the number of iterated detection of unknown images and accelerates the speed of image matching.