基于压缩感知的图像检索方法研究

针对大尺寸图像的特征提取算法复杂度高、特征信息容易缺失的问题,利用压缩感知理论中关于少量测量值可以精确重构原始信号的特性,提出了一种基于压缩感知的图像检索方法。首先对图像进行小波变换、分块预处理;然后构造分块多项式确定性测量矩阵,并对分块图像进行压缩感知快速测量,得到少量的压缩测量值代表图像的特征;最后采用加权距离方法计算图像测量值特征的相似度,实现图像的精确检索。仿真结果表明,该方法在图像检索速度和查准率、查全率等指标上具有更高的性能,能应用于大量复杂图像的检索。

An Image Retrieval Method Based on Compressive Sensing

For solving the problems of the complexity about feature extraction on large size image and the loss of feature information, the characteristics of compressive sensing (CS) theory that a small amount of measurements can accurately reconstruct the original signal is used, and a new image retrieval method based on compressive sensing is proposed. First, the wavelet transformation is performed,and the image is divided into blocks. Then,blocked polynomial deterministic measurement matrix and conduct fast CS measurement on blocked image constructed, and very few compressive measurements which represent the image features can be obtained. Finally, we calculate the similarity between the measurement features by weight distance is calculated, so to implement accurate image retrieval. At the same time, it is proved theoretically that the blocked polynomial deterministic measurement matrix is satisfied to the restricted isometric property (RIP). Experimental results show that this method has higher performance on image precision and image recall, and can be applied to massive complex image retrieval.