基于图像内容的服装分类和推荐方法研究

发布时间：2018-11-25 12:13

【摘要】：随着服装产业的飞速发展,通过计算机视觉技术解决服装领域相关问题已经成为近年来的研究热点。由于服装款式和服装风格的多样性与复杂性很大程度上增加了个性化服装推荐的难度,使得目前所拥有的服装推荐的准确率仍有待提高。针对如何实现能够推荐最适合用户肤色、发色、瞳色及风格的服装,本文提出基于图像内容的服装分类和推荐方法。针对服装款式和风格的多样性与复杂性,本文提出了一种基于特征提取和距离度量的服装风格分类方法。首先,建立服装图像数据集;然后,通过土耳其机器人对所有服装图像的风格、属性进行标注;之后,对所有图像进行姿势估计和特征提取;最后,采用特征提取和距离度量对多个服装属性标签分类,实现基于服装风格的服装分类。针对已有的服装推荐算法中服装色调与用户颜色特征不协调以及推荐的准确率不高的问题,将服装搭配的四季色彩理论与计算机视觉领域的推荐方法相结合,本文还根据该风格分类方法的分类结果提出一种基于图像内容的服装推荐方法。首先,提出四季色彩判断模型,对根据输入的人脸图像提取得到的用户颜色特征集进行分类;然后,建立优化处理模型,根据四季色彩判断模型结果和用户所需风格进行优化处理并获得服装预推荐结果;最后,通过用户评分及反馈机制,提高优化结果,得到最终推荐结果。实验表明,本文所提出的基于特征提取和距离度量的服装风格分类方法可以有效地对不同风格的服装进行分类。提出的基于图像内容的服装推荐方法能够实现与用户颜色特征相协调,并且在实际应用中其推荐结果具有较高的准确率。
[Abstract]:With the rapid development of garment industry, it has become a research focus in recent years to solve the related problems in clothing field by computer vision technology. Due to the diversity and complexity of fashion styles and styles, the difficulty of personalized clothing recommendation is greatly increased, so the accuracy of clothing recommendation still needs to be improved. In view of how to recommend the most suitable color, hair color, pupil color and style of clothing, this paper proposes a method of clothing classification and recommendation based on image content. In view of the diversity and complexity of fashion styles and styles, this paper presents a method of fashion style classification based on feature extraction and distance measurement. First, the clothing image data set is established; then, the style and attributes of all the clothing images are annotated by the Turkish robot; after that, the pose estimation and feature extraction of all the images are carried out. Finally, feature extraction and distance measurement are used to classify multiple clothing attribute labels to achieve clothing classification based on clothing style. In order to solve the problem that the color features of clothing color and user color are not in harmony and the accuracy of recommendation is not high, the four seasons color theory of clothing collocation is combined with the recommended method in the field of computer vision. According to the classification results of the style classification method, this paper proposes a clothing recommendation method based on image content. Firstly, the four seasons color judgment model is proposed to classify the user color feature set extracted from the input face image. Then, the optimization processing model is established, according to the results of the four seasons color judgment model and the style required by the user, the optimal processing is carried out and the clothing pre-recommendation result is obtained. Finally, through the user rating and feedback mechanism, improve the optimization results and get the final recommendation results. Experimental results show that the proposed method based on feature extraction and distance measurement can effectively classify different styles of clothing. The proposed clothing recommendation method based on image content can coordinate with the color features of users and has a high accuracy in practical application.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP391.3

【参考文献】