基于距离度量学习和多视图学习的服装主观风格识别方法

发布时间：2018-07-09 10:13

本文选题：服装主观风格 + 距离度量学习　；参考：《浙江大学》2016年硕士论文

【摘要】：随着电子商务的发展,如何正确识别服装风格成为一项意义深远的工作。服装风格的准确识别可为如下研究提供基础：自动化的服装标注、基于内容的服装检索、个性化的服装推荐、风格统一的服装搭配。而现有服装风格识别方法主要基于客观风格,较少基于主观风格,且缺乏对多个主观风格同时识别的工作,因此本文提出一种基于多个主观风格的服装识别方法。我们首先从互联网上爬取多张带有多个主观风格的服装图片及其主、客观风格信息；然后对这些图片进行姿势估计和特征提取；其次为了能够得到更有区分度的特征,我们提出一种多标签距离度量学习模型,并将该模型运用到已获得的特征上,得到新的距离度量学习特征。之后为了进一步提高服装主观风格的识别效果,我们结合服装图片的客观风格文本描述信息,将已有的单标签多视图学习模型改进成为多标签多视图学习模型,得到多视图学习的新特征。最后对多视图学习特征集进行多标签分类,得到服装多个主观风格的识别结果。实验结果表明,本文提出的基于距离度量学习和多视图学习的服装主观风格识别方法可以有效地识别服装主观风格。
[Abstract]:With the development of e-commerce, how to correctly identify clothing style has become a far-reaching work. The accurate identification of clothing style can provide the basis for the following research: automatic clothing labeling, content based clothing retrieval, personalized clothing recommendation, uniform style of clothing collocation. However, the existing fashion style recognition methods are mainly based on objective style, less on subjective style, and lack of recognition of multiple subjective styles at the same time. Therefore, this paper proposes a multi-subjective style based clothing recognition method. First, we crawl a number of clothing images with multiple subjective styles and their subjective and objective style information from the Internet; then we estimate the posture of these images and extract the features; secondly, we can get more differentiated features. We propose a multi-label distance metric learning model and apply the model to the acquired features to obtain a new distance metric learning feature. Then, in order to improve the recognition effect of clothing subjective style, we improve the existing single-label multi-view learning model into multi-label multi-view learning model by combining the objective style text description information of clothing image. A new feature of multi-view learning is obtained. Finally, multi-view learning feature sets are classified by multi-label classification, and the recognition results of multiple subjective styles of clothing are obtained. The experimental results show that the proposed method based on distance metric learning and multi-view learning can effectively identify the clothing subjective style.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TS941.2

【相似文献】