SIFT 算法被广泛应用于图像特征提取与匹配。由于在利用 SIFT 特征进行图像匹配时,需要计算128维 SIFT 描述子间的欧氏距离,这对于大规模的图像检索耗费时间巨大。针对上述问题,提出一种利用二值 SIFT 描述子(EBSIFT)进行图像匹配的方法。首先,将128维 SIFT 描述子隔点作差,将差值与阈值的比较结果用2位二进制数表示,获得256维二值 SIFT 描述子;然后,将128维 SIFT 描述子隔点求平均值,获得128维均值 SIFT 描述子,再按上述同样的方法对这128维均值 SIFT 描述子隔点作差,再次获得256维二值 SIFT 描述子,从而获得512维联合二值 SIFT 描述子;最后,在进行图像匹配时,利用汉明距离计算512维二值 SIFT 描述子间的距离。实验结果表明,该方法的匹配正确率达到99.58%,与原 SIFT 算法持平,而匹配速度是原 SIFT 算法的19倍,大幅提高匹配效率。%SIFT algorithm has been widely used in image feature extraction and matching.However,the Euclidean distance between 128-dimensional SIFT descriptors needs to be calculated while executing image matching based on SIFT features,which is a time consuming process for large scale image retrieval.To solve the above problem,an image matching method using binary SIFT descriptor (EBSIFT)is proposed.Firstly,through calculating the interval dot difference of 128-dimensional SIFT descriptor and representing the comparison result of difference and threshold with 2 bit binary number,the 256-dimensional binary SIFT descriptor is obtained.Secondly,through calculating the interval dot mean value of 128-dimensional SIFT descriptor,the 128-dimensional mean SIFT descriptor is obtained.Then we calculate the interval dot difference of the obtained 128-dimensional mean SIFT descriptor as above,and the 256-dimensional binary SIFT descriptor is obtained again.Thus,the further 512-dimensional joint binary descriptor is obtained.Finally,hamming distance is used to calculate the distance between 512-dimensional binary SIFT descriptors in image matching.Experimental results show that the matching accuracy of the proposed method achieves 99.58%,which is equal to that of the original SIFT algorithm.At the same time,its matching speed is 19 times higher than the original SIFT algorithm,which greatly improves the matching efficiency.
展开▼