【作者】 李炜；

【导师】 李继云；

【作者基本信息】 东华大学 ， 计算机应用技术， 2013， 硕士

【摘要】 随着服装产业的快速发展,服装推荐和检索系统在为消费者挑选满意的服装时扮演着越来越重要的作用。传统的服装推荐系统通过服装图像文本标注或者服装图像内容(如颜色、纹理等)来推荐。近年来,一些研究者开始基于感性工学尝试从人类感知的角度去推荐服装。感性工学确实能在一定程度上用高层语义表达用户的情感需求,然而感性工学的数据是基于大量的问卷调查和统计分析而得到的,其缺乏动态性和个性化。本文以男士衬衫设计为背景,主要研究内容是男士衬衫的情感推荐。针对感性工学推荐结果缺乏动态性和个性化,提出基于情感语义的男士衬衫个性化推荐系统,最后该系统在情感语义衬衫智能设计系统中作为一个模块被实现。首先基于Kobayashi的色彩心理学建立一个通用情感模型,用于对衬衫图像数据库中的所有图像进行自动情感标注,并把这个标注结果作为系统的默认推荐。其次为了满足个体的个性化需求,我们也将形状、纹理等影响情感的因素考虑进来,系统通过与用户交互来获得用户的情感偏好,然后通过个性化引擎来建立衬衫图像底层视觉特征与用户高层情感语义间的映射关系,并以此建立个性化情感模型。实验表明,系统经过与用户2到3轮的交互,能够很好地推荐用户满意的衬衫。针对男士衬衫设计效率不高问题,提出了情感语义衬衫智能DIY设计,最后该方法在情感语义衬衫智能设计系统中得以实现。在系统中用户能够用高层情感语义来表达需求,并能根据用户情感输入快速推荐相应的部件、颜色和图案,然后用户可以根据部件进行自由拼接。如果用户不满意当前的推荐结果,系统还提供了许多交互式设计功能(如色彩设计、图案设计、液化和纹理滤镜等)来尽可能多的满足用户的个性化情感需求。更多还原

【Abstract】 With the rapid development of garment industry, clothing recommendation and retrieval system plays a more and more important role in choosing satisfied garments for consumers. Traditionally these systems only recommend by either clothing image text annotation or clothing image content, e.g. colors etc. Recently, some researchers start trying to recommend clothing from human sensibility based on Kansei Engineering. Kansei can express user’s emotional demand in high semantic level. However, Kansei Engineering methods suffer from lacking of dynamic support and result personalization due to the fact that Kansei data are collected from a large number of fixed predesigned questionnaires and the final conclusion is drawn by statistic calculation of those data. The main researched content of this paper is about the shirt recommendations in the background of shirt design.To tackle the lack of personalization, in this paper, we propose a personalized shirt recommendation system(PSRS abbr.) based on dynamic mapping between the low level features’ space of shirt images and the high level user’s emotional semantic space. Finally, the PSRS is implemented as a module of system that named shirt intelligence design system based on emotional semantics(SIDS abbr.).We build a common user emotion model based on Kobayashi’s color psychology theory, and automatically annotate all shirt images in database with this emotion model. Then we utilize the result of emotional annotation as the default recommendation of our system. To meet the personalized requirement of each individual and also take other affective factors of clothing choosing such as shape, texture etc. into consideration in the mean time, we collect and update user’s preference through interaction with the user, and then map from the low features space of the shirt image into the user’s high level emotional semantic space by Personalized Recommender engine. Thus we build our personalized user emotion model. Experiments show that by two or three rounds of interaction, our system’s recommendation can meet the personalized emotional requirement of the user well.To tackle the low efficiency of shirt design, we proposal the shirt intelligence DIY design based on emotional semantics, which is implemented in SIDS. In SIDS, user can freely splice and design shirts with the materials(such as shirt parts,color, pattern) which are recommended by the system according user’s high level emotional adjectives input. If the user don’t satisfy the current recommendations, he can use the interactive design tools(such as color design, pattern design, Liquify and Texturizer tools, etc.) to personalize his requirements as much as possible.更多还原