一种基于多特征签名的图像检索系统

A MULTI FEATURE SIGNATURE BASED IMAGE RETRIEVAL SYSTEM

针对以浮点矢量形式保存的图像特征存储开销大、距离计算复杂的缺点,提出了一种基于多特征签名的图像检索系统。该系统利用主分量分析和矢量量化技术,对多类浮点矢量特征降维后映射到多个特征签名中,并通过汉明距离表示特征签名之间的距离。实验结果表明,该系统相对于基于浮点矢量的图像检索系统能很好地实现图像的特征存储和签名匹配,在准确率保持不变的情况下能返回更多的检索结果,且具有较好的特征可扩展性。

Aiming at the huge storage cost and the complexity of distance computation for image features stored in float vector format,a multi feature signature based image retrieval system has been put forward.By using Principal Component Analysis（PCA） and Vector Quantization（VQ）,the system reduces the dimension of multiple float feature vectors that are reflected onto multiple feature signatures.Then Hamming Distance is introduced to represent the distance between feature signatures.Experiments demonstrate that,compared to float vector based image retrieval systems,the novel proposed system can handle feature storage and signature matching issues,returning more query results without downgrading the correctness ratio.Moreover,the system has excellent feature extensibility.