摘要
检索图像时,由于高频信息的分析精度较低,导致检索性能偏低,为此提出基于特征聚类的网络海量图像快速检索方法。应用4个一阶梯度算子和1个二阶梯度算子提取图像的高频信息,并计算均值曲率,在均值曲率参数差距小于残差值的最大值,且满足平滑值要求的前提下,对图像进行聚类处理。在检索阶段,以图像的均值曲率特征为基础,确定对应的检索聚类范围,再将残差值和平滑值作为具体检索结果的输出基准。在测试中,设计方法检索结果的归一化折损累计增益与数据集标签种类、测试图像质量之间未表现出明显的相关关系,始终稳定在0.95以上,具有良好的巡回性和较高的检索效率。
When retrieving images,due to the low accuracy of high-frequency information analysis,the retrieval performance is low.Therefore,a feature clustering based network massive image fast retrieval method is proposed.Apply four first step operators and one second step operator to extract high-frequency information from the image,calculate mean curvature,and cluster the image on the premise that the difference in mean curvature parameters is less than the maximum residual value and meets the smoothness requirements.In the retrieval stage,based on the mean curvature feature of the image,the corresponding retrieval clustering range is determined,and then the residual value and smoothing value are used as the output benchmark for the specific retrieval results.In the testing,there was no significant correlation between the normalized cumulative gain of the design method retrieval results and the type of dataset labels and the quality of the test image,and it remained stable above 0.95.It has good roundness and high retrieval efficiency.
作者
郭英英
GUO Yingying(Haojing College of Shaanxi University of Science&Technology,Xi'an Shaanxi 712046,China)
出处
《信息与电脑》
2023年第17期48-50,共3页
Information & Computer
基金
2021年陕西省教育厅科研项目“基于本体的青花瓷多媒体信息系统研究”(项目编号:21JK0554)。
关键词
特征聚类
海量图像
快速检索
feature clustering
massive images
quick retrieval
