Research Of Image Matching Algorithm Based On Local Features

Image matching technology has always been one of the hot research directions in the field of image information processing and has been widely used in military,industry,transportation and many other fields.In recent years,human-computer interaction teachnology represented by computer vision technology and artificial intelligence technology is booming.Image matching is a basic technology required for human-computer interaction teachnology research.So the research of image matching technology has important theoretical and practical significance.Image matching algorithms can be roughly divided into the following three categories: the image matching algorithms based on gray correlation,based on information in the transform domain and based on feature.The image matching algorithms based on local feature are characterized by its relatively small amount of computation and strong robustness,which has become the mainstream research direction.The Image features are divided into global features and local features.In this thesis,the main work of image matching algorithms based on local feature is as follows:(1)Aiming at the poor matching ability of the scale-invariant feature transform(SIFT)algorithm for color images with different light intensities,an improved color image matching algorithm based on SIFT is proposed in this dissertation.The improvements are as follows: Firstly,every SIFT feature descriptor is reduced in dimension from 128 to 36,and the color information of every feature point and the pixels within its neighborhood is incorporated into the corresponding 36-dimensional SIFT feature descriptior inorder to generate a 42-dimensional feature descriptior containing color information.Secondly,the preliminary matching and the exact matching of the color image are completed in order.Lastly,a method based on the relative distance theory is adopted to eliminate the mismatching points.The simulation results indicate that under the condition of varying illumination intensity,compared with the classic SIFT algorithm,the algorithm can effectively improve the accuracy of color image matching.Compared with some common SIFT improved algorithms,it can effectively improve the efficiency of color image matching.(2)To solove the problem that SIFT algorithm used to match the image containing a large moutain of similar aera has poor local image matching performance,this thesis proposes a SIFT algorithm based on Contourlet transform.Firstly,the SIFT algorithm is used to obtain the DoG spatial extremal points.Secondly,an adaptive threshold Canny edge detection algorithm is used to detect the edge of the image to remove the edge points from the DoG spatial extremal points.Lastly,the global texture information extracted by the Contourlet transform and the local feature information extracted by the SIFT algorithm are comprehensively used for image matching.The simulation results show that compared with SIFT algorithm,the algorithm has higher matching accuracy.(3)Aiming at the Weak Rotation Robustness of SURF Algorithm,a LBP-SURF image matching algorithm is proposed in this thesis.The algorithm firstly incorporates the information extracted by an improved rotation invariant LBP feature description method into the SURF feature descriptiors.Secondly,instead of euclidean distance,the number multiplication distance is used for feature matching.The simulation results show that the algorithm has better rotation invariance than SURF Algorithm while ensuring certain the real-time.