Study On Image Matching Algorithm Based On Improved SIFT

Image matching is a fundamental but one of the most important issues in computer vision.It has been widely used in target tracking,space exploration,3D reconstruction,modification identification,and so on.In recent years,quite a number of algorithms have been proposed,and significant progress has been made in the field of image matching.Currently,there are three basic classes of image matching algorithms: gray-value matching,feature matching and transform-domain matching.Feature matching algorithms have been widely studied due to their low complexity and high robustness.In general,local descriptors are used to characterize the image region around feature points in feature matching algorithms.The methods of detecting feature points include Gaussian derivatives,Harris corner detector,moment invariants,shape context,SIFT,speeded up robust features(SURF)and so on.In particular,the SIFT descriptor,which remains invariant to rotation,scale or illumination changes,is also robust to vision changes,affine changes or noise,and has an outstanding performance for feature matching based on feature points.However,in order to reduce the complexity,the first step of SIFT algorithm is to convert the input color images into grayscale images,in which the feature points are detected.Therefore,it does not take the color information available in the color images into account.Because of the loss of color information,the contrast of the color images will be decreased,which leads to the results that the feature points will not be detected in the regions which have a similar grayscale level but different hues.The loss of color information may also result in decreasing of matching ratio.A great deal of work is done to study image matching algorithms based on SIFT in this thesis.First of all,the background,significance and the research status quo of image matching technology are briefly introduced.Secondly,the principle,key elements,general flow and classification of image matching are described in detail,and several classic matching algorithms are introduced.Thirdly,the image matching algorithm based on SIFT is presented,and the flowchart of algorithm is given.In the meantime,the extraction and matching process of SIFT feature descriptor are introduced in detail and the advantages and existing problems of thealgorithm are analyzed.Finally,two improved algorithms are proposed to solve the problem of SIFT algorithm in loss of color information:(1)Image matching algorithm based on SIFT using image grayscale: The image gray scale algorithm is applied to the first step of the SIFT algorithm.Firstly,the color images are converted from RGB color space to CIE L*a*b color space.Secondly,the chrominance differences between adjacent pixels are calculated.Finally,the chrominance differences are iteratively optimized to get the final grayscale.(2)Image matching algorithm based on SIFT using color and exposure information: Through the analysis of color images,the color information and the exposure information,which are calculated,are added to the original grayscale images in order to enhance the contrast of the regions in the grayscale.In the meantime,the exposure is adjusted.Both improved algorithm and original algorithm are simulated using multiple sets of images.The experimental results show that both improved algorithm can effectively differentiate the regions with different colors but the similar grayscale level,and increase the matching ratio of image matching based on SIFT.