An Image Matching Method Based On MBR-SIFT Descriptors

Image matching is a critical aspect for image analysis technology related tasks in image retrieval,image categorization and object recognition.The scale invariant feature transform(SIFT)method has been widely applied in image matching.In the matching procedure,the 128-D descriptors of all keypoints in two images are extracted.The 128-D descriptors of each keypoint of the first image are compared to those of the second one.The Euclidean distance is used as the similarity measurement of two descriptors to find the nearest matching keypoint.SIFT algorithm usually generates hundreds to thousands of keypoints for each image.And correspondingly,the SIFT features could be numerous in a large image database.Moreover,the distance computation involves taking square root.Therefore,image matching in the SIFT method to large-scale image database is highly time-consuming.To address the aforementioned problems,many binary SIFT(BSIFT)methods have been developed to improve matching efficiency.However,both of SIFT and BSIFT methods cannot provide a solution to the image matching problem of mirror reflection.Therefore,this paper mainly includes the following work according to the current research status:1.To solve the time consuming problem,this paper proposes a novel BSIFT method.Specifically,this method compares the difference values of a 128-dimensional(128-D)descriptor and a threshold;the comparison results are denoted by 2-bit binary numbers,thereby converting the 128-D descriptor into a 256-bit binary SIFT descriptor.In addition,the linear relationship between the threshold and the standard deviation of the 128-D descriptor is employed to derive an equation for the threshold.Furthermore,to avoid the error result caused by Hamming distance,this paper redefines the distance measure of 256-bit BSIFT descriptors.The experimental results show that the proposed method can speed up the matching and ensure the matching accuracy.2.Both of SIFT and BSIFT methods are not invariant to mirror reflection.In order to address this issue,this paper presents a novel descriptor which is horizontal or vertical mirror reflection invariant,named MBR-SIFT.First,16 cells in the local region around the SIFT keypoint are reorganized,and then the 128-dimensional vector of the SIFT descriptor is transformed into a reconstructed vector according to eight directions.Then,the reconstructed descriptors are binarized and executed the reverse coding.The proposed descriptor is not only invariant to horizontal or vertical mirror reflection,but also invariant to rotation,scale,perspective,illumination and blurring.3.Based on the MBR-SIFT descriptor which is invariant to mirror reflection,this paper proposes an image matching method with coarse and fine selection combining with binarization and similarity measure methods.By comparing with BRIEF,BRISK,FREAK,SIFT,Chen's and Zhou's,the experimental results on Oxford and UKBench data sets show that the proposed method is invariant to rotation,scale,perspective,illumination and mirror reflection and ensures higher efficiency.