基于深度残差网络的服装图像检索

Clothing Image Retrieval Based on Deep Residual Network

针对服装图像检索这一问题,用深度残差网络ResNet101作为骨干模型,并使用DeepFashion数据集中的子数据集Category and Attribute Prediction Benchmark中的图片作为研究对象。文中首先将服装图片进行处理后送入已经训练好的网络模型中得到服装图像特征,并建立服装特征库,然后将待检索的服装图片送入模型中得到图像特征,并与服装特征库中的特征进行相似度度量,最后按照相似度大小得到检索结果。实验结果表明,该方法可以得到比较完整的服装特征,具有较高的检索准确率。

In order to solve the problem of clothing image retrieval,the deep residual network Resnet101 is used as the backbone model,and the images in the Category and Attribute Prediction Benchmark,a sub-dataset of the DeepFashion dataset,are used as the research object.In this paper,first of all be incorporated into clothing pictures after processing has trained network model in the clothing image characteristics,and establish the clothing characteristic library,and then to retrieve the clothing pictures into the model of image characteristics,and with the clothing features in the library for a similarity measure,finally according to the size of similarity retrieval results are obtained.Experimental results show that this method can obtain more complete clothing features and has a high retrieval accuracy.