结合组SIFT特征和多元词袋模型的图像检索方法

摘要

针对现有图像检索方法召回率低的问题,提出一种结合组尺度不变特征变换(SIFT)特征和多元词袋模型的图像检索方法.采用间隔采样方式抽取图像中的关键点,提取SIFT特征;依据图像的空间相关性,将各关键点及8邻接关键点的SIFT特征进行组合,构建组SIFT特征;针对组SIFT特征,在经典词袋模型的基础上,加入Dirichlet函数和组混合编码,构建多元词袋模型,提取图像特征向量;采用卡方距离度量和非对称距离计算方法快速计算特征向量之间的相似度,得到图像检索结果.仿真结果表明,该方法的召回率指标明显高于对比方法,查准率指标也有提高.%Facing the problem of low recall performance of existing image retrieval methods,an image retrieval method combining with group scale-invariant feature transform (SIFT) features and multi-bag-of-words model was proposed.Key points in an image were selected using interval sampling mode,and SIFT features were extracted for them.SIFT features of each key point were combined with its 8-adjacent key points according to spatial correlation of images,and group SIFT features were built.On the basis of classic bag-of-words model,Dirichlet function and group mixing coding were added for these group SIFT features,and multi-bag-of-words model was built for extracting feature vector from an image.Chi-square distance metric and asymmetric distance calculation method were used for fast computing the similarity between two feature vectors,and results of image retrieval were obtained.Simulation results show that,the recall value of the proposed method is significantly higher than the comparative methods,and the precision value is also improved.