Research On Curve Matching Technology Based On Descriptor

Curve matching technology is a key and widely used technology in pattern recognition.It is also widely used in practical applications such as 3D reconstruction,visual navigation,heritage restoration,video surveillance system,robot technology and image stitching.The process of curve matching technology is divided into the following three steps:the curve is extracted by matching the image with a certain method;the curve is characterized by the corresponding algorithm;the similarity between the two curves is calculated by matching criterion and the matching criterion is used to achieve match.In recent years,although the research on feature-based curve matching technology based on descriptors has made great progress,but with the widespread use of mobile devices,it is necessary to study binary curve descriptor with small storage space and fast matching speed.At the same time,the extracted characteristic curve is affected by the following factors:the end of the curve can't be accurately controlled at both ends of the curve,the proposed characteristic curve length is not the same,the curve contains more repetitive texture,etc..These factors lead to the study of curve matching based on texture feature is still a hotspot and difficult research.Therefore,this thesis gives deep content study,the main innovations of the thesis are:(1)Since the binary descriptor has the advantage of low storage and fast matching,we use the threshold method to convert the curve descriptor into the binary code string composed of 0 and 1 on the basis of real type curve descriptor study(MSCD?IOCD?IOMSD?TCHP).And obtain two binary curve descriptors:Binary Descriptor based on Dichotomy and Binary Descriptor based on Triple Value Method.The threshold of the former is the mean of the real-type curve description vector with self-adaptability,which is binarized in the original dimension of the real-type curve descriptor(except for the B~1-IOMSD descriptor)to obtain the binary descriptor composed of 0,1,and its storage space is reduced to 1/32 of the real number curve descriptor.The latter consists of two adaptive thresholds:the sum of the mean and the t-times the mean square and the difference,and the real-type curve description vector is binarized to obtain the binary descriptor composed of00,01,11,the storage space Reduced to 1/16 of the real type curve descriptor.The experimental results show that the binary curve descriptor improves the matching accuracy of the descriptor under the condition of keeping the real-type curve descriptor's invariant rotation invariance,the change of illumination invariance and the deformation stability.Reducing the memory consumption of the descriptor greatly under the condition of increasing the time consuming.(2)For the shadow problem of the image,the curve descriptor based on the brightness order will be affected by it to increase the false matching of the image.This paper presents a novel curve matching algorithm:Gradient Order Curve Descriptor(GOCD).Firstly,the support region of the curve is determined,and then the pixels with the smaller gradient magnitude in the support domain are removed and each curve support region is divided into several sub-regions according to the gradient magnitude order of the remaining pixels.Finally,each sub-region descriptor is constructed by using the gradient order feature to obtain the curve Descriptor.Compared with the descriptor based on the brightness order,the gradient order used in the algorithm has stronger robustness to the illumination change,and can reduce the false matching under the shadow of the image,thus improving the resolution of the descriptor.The experimental results show that the proposed algorithm has better matching performance in terms of matching performance,such as viewpoint change,JPEG compression,illumination change,wide baseline,rotation and noise image than the intensity-based order descriptor.Experiments show that the two proposed texture-based curve descriptors(real-type curve binary descriptors,GOCD)have a good effect on the matching performance and matching accuracy for a variety of image changes,which can achieve the automatic matching of curves in the image,and the former also greatly reduces the storage space for image description features.