一种海量图像下的高精度特征检索算法

A High-Precision Feature Retrieval Algorithm for Massive Images

传统图像关键特征二次检索算法无法计算图像纹理相似度,导致数据库中海量相似图像的检索精度偏低,查全率与查准率均不理想。为此,设计新的基于灰度直方图的图像关键特征二次检索算法。采用中值滤波将某像素点灰度值替换成相邻区域像素点灰度中值,去除图像背景及噪声干扰。建立归一化灰度共生矩阵,并向量化输入图像并计算其灰度特征矢量。利用灰度直方图函数计算图像纹理相似度,完成图像关键特征第一次检索。运用Canny算子检测图像边缘并统计各边缘长短,计算输入图像与待检测图像投影值,获得图像关键特征形态信息,实现高精度图像关键特征二次检索。仿真结果证明,所提方法的查全率与查准率均较高,能快速精准检索出大量相似的图像特征。

Currently, the retrieval of massive similar images in the database has low accuracy, recall and precision. Due to the lack of calculation of image texture similarity in the secondary retrieval algorithm of image key features. Therefore, this paper designs a secondary retrieval algorithm of image key features based on the gray histogram.Firstly, the median filter was introduced to replace the gray value of a pixel with the gray value of pixels in adjacent areas for deleting the image background and eliminating noise interference. Secondly, the normalized gray level cooccurrence matrix was established, the input image was vectorized and its gray level feature vector was calculated.Then, the image texture similarity was calculated via the gray histogram function, completing the first retrieval of the key features of the image. Then, the Canny operator was used to detect image edges and count the length of each edge. The projection values of the input image and the image to be detected were also calculated to obtain the morphological information of image key features. Eventually, the secondary retrieval of high-precision image key features was completed. The simulation results show that the recall and precision of this method are better than those of traditional algorithms.