基于深度学习与拓展查询的商标图像检索方法

摘要

In order to solve the problem of "semantic gap" in the retrieval of trademark images,this paper presents a method of trademark image retrieval based on depth learning.Because of the excellent performance of depth learning technology in image recognition and image classification,we use it to solve a series of problems encountered in trademark image retrieval.This paper also improves the query expansion algorithm and uses it to optimize the retrieval system.First of all,we construct a trademark image data set using the trademark image.Secondly,the trademark image library is used to train a stable Alex Net convolution neural network model and construct the trademark Image feature library.Then,we use the Training network model to extract the deep semantic features of the imported trademarks and measure the similarity by the distance function.Finally,we propose an improved query expansion method,which is used to expand the candidate image and get similar images.The experimental results show that the method of trademark image retrieval based on depth learning has better retrieval effect than the traditional method based on SVM active learning.%为了解决商标图像检索时,由于图像抽象性和复杂性的特点所带来检索时的“语义鸿沟”问题,提出一种基于深度学习的商标图像检索方法.由于深度学习技术在图像识别和图像分类技术中有着优秀表现,因此我们利用它来解决商标图像检索时遇到的一系列问题.本文还改进了拓展查询算法,并利用它优化检索系统.首先,我们利用商标图像的特征构建了一个多类别的商标图像数据集.其次,利用商标图像库训练一个稳定的AlexNet卷积神经网络模型,并构建商标图像特征库.接着,利用训练好的网络模型提取输入商标的深层语义特征并通过距离函数进行相似性度量.最后,提出一种改进的拓展查询方法,利用其对候选图像进行拓展查询,得到相似图像.经过实验结果表明,基于深度学习的商标图像检索方法比传统基于SVM主动学习的图像检索方法有着更好的检索效果.