基于内容的商标图像检索研究

发布时间：2018-03-16 00:05

本文选题：商标图像检索　切入点：形状特征　出处：《华南理工大学》2010年硕士论文　论文类型：学位论文

【摘要】： 目前商标局采用传统基于类目和文本的商标检索方法来对待注册的商标进行审查,这种基于人工分类和关键词标注的方法,不仅主观,且耗费大量的人力和时间。受到商标局人员编制的限制,采用传统的商标检索方法使得商标审查能力相对滞后,造成商标注册申请大量积压,不利于市场经济的飞速发展。近年来发展的基于内容的图像检索(CBIR)方法突破了这种传统方法的弊端,它利用图像自身的特征(颜色、纹理、形状、空间结构信息、语义等)实现对图像的自动、高效检索。国内外很多研究人员已经把CBIR技术应用到商标图像检索上,以缩短商标审查时间,提高审批效率。本文在对商标图像特点和基于内容的商标图像检索相关技术进行一定研究分析后,提出了一种综合图像多特征的检索方法,设计实现了一个商标图像检索原型系统。首先对基于内容的商标图像检索进行了综述,分析了基于内容的图像检索中所涉及的相关技术,包括特征提取技术、图像匹配技术、检索结果评价技术等。针对商标图像的特征,重点研究了常用的形状特征提取方法,如形状不变矩,边界方向直方图、偏心率等。并介绍了四叉树分块和基于极坐标分块两种商标图像检索方法。针对全局特征和局部特征单独检索商标的不足,提出了一种综合多特征的商标图像检索方法,该方法首先提取图像的偏心率、7个HU不变矩、边界方向直方图,加权组合这三种特征进行初次检索,接着采用基于极坐标的分块方法提取分块图像特征,在初级检索的结果集上进行二次检索得到最后的检索结果。最后,设计实现了一个综合多特征的商标图像检索原型系统,该系统包括人机交互、图像预处理、特征提取、图像检索等模块。在这个系统上对商标进行检索,并通过这个系统平台,验证本文提出算法的有效性。
[Abstract]:At present, the Trademark Office uses traditional trademark retrieval methods based on category and text to review registered trademarks. This method based on manual classification and keyword marking is not only subjective. Due to the limitation of the Trademark Office's staffing, the traditional trademark retrieval method makes the ability of trademark review lag behind, resulting in a large backlog of trademark registration applications. The CBIR-based image retrieval method developed in recent years has broken through the drawbacks of this traditional method, which utilizes the features of the image itself (color, texture, shape, spatial structure information), which is not conducive to the rapid development of the market economy. Many researchers at home and abroad have applied CBIR technology to trademark image retrieval in order to shorten the time of trademark review and improve the efficiency of examination and approval. In this paper, the characteristics of trademark image and the related techniques of content-based trademark image retrieval are studied and analyzed, and a multi-feature image retrieval method is proposed, and a trademark image retrieval prototype system is designed and implemented. Firstly, this paper summarizes the content-based trademark image retrieval, and analyzes the related technologies involved in content-based image retrieval, including feature extraction and image matching. According to the characteristics of trademark image, the common methods of shape feature extraction, such as shape invariant moment, boundary direction histogram, etc. Eccentricity and so on. Two trademark image retrieval methods based on polar coordinate block and quadtree partitioning are introduced. In view of the shortcomings of global and local features in trademark retrieval alone, a multi-feature image retrieval method is proposed. The method firstly extracts the eccentricity of the image, seven Hu invariant moments, and the boundary direction histogram. The first retrieval is carried out by weighted combination of these three features, and then the segmented image features are extracted by using the block method based on polar coordinates, and the final retrieval results are obtained by quadratic retrieval on the primary retrieval result set. Finally, a multi-feature trademark image retrieval prototype system is designed and implemented. The system includes human-computer interaction, image preprocessing, feature extraction, image retrieval and so on. The effectiveness of the proposed algorithm is verified by this system platform.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：TP391.41

【相似文献】