Research On Image Matching Algorithm Based On Content Feature

In the ever-changing information technology today,the single image size and the total number of images is in the geometric multiple of the growth,which will bring great challenges of image management and image retrieval.How to retrieve the image from the massive image is a difficult problem in the field of scientific research,and the content-based image retrieval method is an important means to solve such problem.In this paper,content-based image retrieval and matching is the research direction.The classical SIFT feature extraction operator is used to segment and match the image,then the influence of image noise on SIFT operator is analyzed.The main works of this paper are as followed:In the process of image segmentation,an improved Graph Cuts(GC)image segmentation algorithm based on SIFT algorithm is proposed to manually select the foreground and background scene conditions.The image segmentation algorithm is applied to automatically find the image Segmented seed points.The simulation results show that the improved GC algorithm has better segmentation effect.In the process of image matching,a SIFT feature extraction algorithm based on color histogram is proposed because of the congenital deficiency of traditional SIFT color feature extraction.Firstly,the scale and direction of the feature points is calculated based on the traditional SIFT feature extraction algorithm.Secondly,the calculation of color histogram is increased by the center of the feature point and the window of the 16?16 field.The color histogram is used to reduce the image search range at the first time.Finally,the second match is performed according to the operator of the traditional SIFT feature point,and the matching images are arranged from high to low by similarity.The experimental results show that the improved SIFT algorithm has high precision and recall rate.In the analysis of the influence of image noise on the SIFT algorithm,an improved filtering algorithm is proposed to reduce the influence of impulsive noise on SIFT feature extraction.Firstly,the left and right two peaks of the image histogram are used to identify suspicious noises and determine those coordinates after detecting the image histogram.Then,recursive weighted average in the local window are used to replace suspicious noises with the similarity of neighbor nood pixels.The experimental result shows that The improved filter median filter algorithm has better effect on impulse noise reduction and is beneficial to SIFT feature extraction.In summary,this paper mainly completed the image segmentation,matching and image noise reduction three aspects of works.The results show that the improved algorithm can improve the accuracy and retrieval rate of the image retrieval system,and achieve the goal of improving the image segmentation and matching accuracy.