摘要
品牌服装图像检索目的是根据用户需求输入图像从而直接查找出与目标图片相似的服装,以满足客户查找需求。服装图像检索步骤分两步,首先利用卷积神经网络模型提取数据集图像特征并保存到数据库,然后提取待检索图像的特征向量并与保存好的的图像特征逐一进行相似度匹配。从深度卷积神经网络模型设计出发,通过卷积核删减方法改进VGG16网络模型并对比分析Inception_v3和原VGG16模型结构,通过实验验证他们的性能表现。实验表明,在获取的品牌服装数据集上,改进的modify_vgg准确率达81.7%,比原VGG16模型准确率提高1.4%,比Inception_v3模型提高3.5%。
The purpose of brand clothing image retrieval is to directly identify the clothing that is similar to the target image according to the user′s requirements.The process of apparel image retrieval consists of two steps.Firstly,the image features of the dataset were extracted by the convolution neural network model and saved to the database,then the features of the image to be retrieved were extracted and the similarity between the features of the image to be retrieved and the saved image features was matched.Improve the VGG16 model by pruning the convolutional kernel from the design of deep convolutional neural network model,analyzed the structure of Inception_v3 and the original VGG16 model,and verified its performance through experiments.The experiments show that the accuracy of the improved modify_vgg model is 81.7%on the brand clothing dataset,which is 1.4%higher than the original VGG16 model,and 3.5%higher than that of the Inception_v3 model.
作者
董访访
张明
邓星
DONG Fang-fang;ZHANG Ming;DENG Xing(College of Computer Science,Jiangsu University of Science and Technology,Zhenjiang 212100,China)
出处
《软件导刊》
2022年第8期144-149,共6页
Software Guide
基金
国家自然科学基金青年科学基金项目(61902158)。