摘要
采用当前方法检索图像中存在的特征时,检索特征所用的时间较长,检索得到的特征数量较少,存在检索效率低和召回率低的问题。提出基于堆叠乘积量化的图像特征反馈性检索方法,结合乘积量化算法和加法量化算法得到堆叠乘积量化算法,通过堆叠乘积量化算法对图像进行降维处理,去除图像中存在的冗余信息和无用数据。在亮度、色彩和梯度三个方面对降维处理后的图像进行检索,获得图像的亮度特征、色彩特征和梯度特征,实现图像特征的反馈性检索。仿真结果表明,所提方法的检索效率高、召回率高。
At present, the method is time consuming to retrieve the features existing in the image, and only small numbers of features can be retrieved, leading to low retrieval efficiency and low recall rate. In this article, a feedback retrieval method of image feature based on stacked product quantization was put forward. At first, the product quantization algorithm was combined with the addition quantization algorithm to form the product quantization algorithm. Then, the stacked product quantization algorithm was used to reduce the dimension of image, so as to remove redundant information and useless data in image. Moreover, the image after dimensionality reduction was retrieved from three aspects:brightness, color and gradient, and then the brightness feature, color feature and gradient feature of image were obtained. Finally, the feedback retrieval for image features was achieved. Simulation results show that the proposed method has high retrieval efficiency and high recall rate.
作者
何青
孙红霞
HE Qing;SUN Hong-xia(Hope College,Southwest JiaoTong University,Sichuan Chendu 610400,China)
出处
《计算机仿真》
北大核心
2020年第4期456-459,475,共5页
Computer Simulation
关键词
堆叠乘积量化算法
图像特征
反馈性检索
Stacked product quantization algorithm
Image feature
Feedback search
