无监督随机优化乘积量化图像检索模型

Unsupervised Image Retrieval Based on Random Optimized Product Quantization

随着互联网图像的增多,线性复杂度的最近邻图像检索已很难满足大规模图像检索的性能需求.为了在大规模图像检索任务下保证精度的同时减少检索的时间,提出一种基于乘积量化的近似最近邻图像检索模型.首先通过卷积神经网络初步提取图像特征.然后通过卷积注意力模块对特征进行处理得到增强后的图像特征.接着根据图像本身的语义结构对神经网络进行训练,再通过训练好的神经网络提取图像的语义特征,并使用随机优化乘积量化方法对语义特征进行处理,最终得到与输入图像相似的检索结果最后,通过在大规模数据集NUS-WIDE上与其他模型进行比较分析,实验结果表明所提模型在大数据图像检索时可以提高检索精度,同时降低检索时间.

Along with the increase of Internet images,nearest-neighbor image retrieval with linear complexity has been difficult to meet the performance requirements of large-scale image retrieval.In order to ensure the accuracy and reduce the retrieval time in large-scale image retrieval tasks,an approximate nearest neighbor image retrieval model based on Product Quantization is proposed.Firstly,the image features are extracted and processed through the convolutional block attention module to obtain the enhanced image features.Ac-cording to the semantic structure of the image,the neural network is trained and used to extract the semantic features of the image.The semantic features are processed by the random optimization product quantization method,and the retrieval results similar to the input image are obtained.The experimental results show that the proposed model can improve accuracy and reduce time in large data image retrieval.