基于手绘草图的服装图像检索技术及系统实现

发布时间：2018-03-13 21:11

本文选题：草图检索　切入点：手绘草图　出处：《北京服装学院》2017年硕士论文　论文类型：学位论文

【摘要】：基于互联网时代下多媒体技术的迅猛发展使得数字多媒体的存储信息越来越庞大,因而图像的管理与存储也成为国内外众多学者重要的研究课题。随着长期在该领域的深入探究,基于内容的图像检索技术已获得了显著的科研成果。基于内容与基于文本的检索技术有着较明显的不同之处。基于文本检索的图像检索技术是根据用户输入的关键字来搜索到符合要求的图片。但关键字文本因不同用户主观性而并不一定准确,从而导致检索结果有所偏差。而基于内容的图像检索技术能够较为完善的解决这一问题,不依赖关键字,把用户手绘内容作为输入,根据内容中物体线条整体廓特性进行表现款式的特征提取,提出相应的匹配算法,从而去相似性响应大量图片数据库,达到对图像内容的全面检索,筛查出目标,找到与之相似的图片。于此同时,近年来智慧型手机、平板电脑等触控输入方式的设备在我们的生活中占据了主要位置,使得人机交互形式发生了重大转变。人们在触屏操作的方式下,可以较为准确且轻松通过手绘把想要表达的想法表现出来,进一步通过手绘的线条图在大范围图片库中找到与之近似的像。利用这样的笔式界面技术进行实时检索符合用户的思维与操作习惯,也便于用户描述目标检索图像。基于内容的图像检索在服装领域也运用广泛。经查阅相关资料我们可以得知色彩、图案、样式是服装最具特点的三大特征[1]。而草图是服装外形设计中一类相对高效的信息载体,在目前现有服装款式图的设计流程中,设计完成的服装款式结构是根据款式名称或编号存人入文档。设计师会根据现有或一般的版型基础上设计纸样,由于服装的款式都是可以在基本纸样结构上作相应调整而到达最后的纸样结构,若仅以文字的描述去标注款式缺少直观性,局限性较大,如以图像的形式作搜索条件可减少操作提高效率。因此,为了提高服装设计师设计样式的效率,提高原有传统算法SURF以提高特征匹配检索效率,在此基础上本文设计并完成了基于服装的草图检索系统。基于草图的服装图像检索系统主要难点表现在如下几个要素:改变原有SURF算法使其具有提取具有服装结构描述的特征;草图与样本图像之间特征提取和描述的一致性,使草图特征与样本特征所描述内容相同;大规模数据库检索的效率,在保持大数据基础不变的情况下提升检索速率;改变移动客户端与网络服务端的传输方式,更加高效、便捷。系统创新性则体现在如下几方面:将基于手绘的草图图像检索技术与服装创造性地相结合;突破已有传统手绘草图的特征提取算法,将服装结构加入特征描述中;初步并逐步系统性地建立以服装特征为主的大规模数据库。基于草图的服装图像检索系统(Sketch Based Clothes Image Retrieval,SBCIR)的系统界面设置有一定大小绘图区域,用户可以绘制的草图,并将系统中的结果显示。在参数化设计已被广泛运用到服装生产领域中的趋势下,可以减少服装设计师不必要的大量重复性操作,提高生产效率。
[Abstract]:The rapid development of multimedia technology and Internet era makes the storage of digital multimedia information becomes more and more huge and based on the management and storage of image has become an important research topic of many scholars at home and abroad. With the long term in the field of image retrieval technique has gained significant achievements in scientific research. Based on content and retrieval technology based on the text is different obviously. The text retrieval based on image retrieval technology is based on user input keywords to search to meet the requirements of the picture. But because of the different with the keyword text and headed view is not accurate, resulting in biased search results. Technology can better solve the problem of content image retrieval based on keywords, does not depend on the user, hand-painted contents as input, according to the overall characteristics of the object content in profile lines Feature extraction for performance styles, put forward the corresponding matching algorithm, in order to respond to a number of similar picture database, to achieve a comprehensive retrieval of image content, screening out the target, find a similar picture. At the same time, the intelligent mobile phone tablet computer in recent years, the touch input device occupies the main position in the in our lives, making human-computer interaction form has changed. People in the touch-screen operation mode, can accurately and easily by hand to want to express the idea, further to find similar as in a large range in the picture library through the lines of the sketch. Using the pen interface technology of such the real-time retrieval with thinking and operating habits of the user, the user can also describe the target image retrieval. Also widely used in the field of clothing image retrieval based on content checking related. The data we can know the color, pattern, style is the three characteristics of [1]. clothing and most of the features of the sketch is a relatively efficient carrier of clothing shape design, in the design process of the existing garment drawing in the fashion design is completed according to the style name or number stored into the document designer. According to the design pattern will version of existing or general, because the clothing styles are available in the basic pattern on the corresponding adjustment to the final pattern, if only in the description of the text to indicate lack of intuitive style, limitations, such as in the form of image search can reduce the operation conditions to improve the efficiency. Therefore, in order to improve the efficiency of designer style, improve the traditional SURF algorithm to improve the matching efficiency of retrieval, this paper is designed based on Service The sketch retrieval system. The retrieval system of clothing image sketch the main difficulties in the following: Based on the performance factors change the original SURF algorithm which has the characteristics of extraction clothing structure description; consistency between sketch and sample image feature extraction and description, the sketch features and sample features described in the same content; the efficiency of large-scale database retrieval in a large data base, to maintain unchanged the retrieval rate; change the mode of transmission, mobile client and server network more efficient and convenient. The system innovation is reflected in the following: the retrieval technology and creative hand-painted clothing sketch image based on the combination of traditional features; breakthrough sketch extraction the clothing structure join algorithm, feature description; preliminary system and gradually to establish large-scale data base on garment feature mainly based on. Clothing image retrieval system (Sketch Based Clothes Image a Retrieval, SBCIR) of the system interface settings have a certain size of the drawing area, the user can draw the sketch, and in the system the results show that the parametric design has been widely applied to the field of clothing production in the trend, can reduce unnecessary repetitive operation a fashion designer, improve the production efficiency.

【学位授予单位】：北京服装学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS941;TP391.41

【相似文献】