【作者】 傅光磊；

【导师】 蒋兴浩； 孙锬锋；

【作者基本信息】 上海交通大学 ， 通信与信息系统， 2010， 硕士

【摘要】 随着互联网科技和多媒体技术的不断发展,数字图像的应用已经渗透到社会生活的方方面面。同时计算机科学也在飞速的发展,硬件设备和软件设备在功能和性能方面不断地进步和创新。在这样的背景下,近年来图像理解问题成为了计算机视觉领域中的研究热点之一。所谓图像理解是指通过设计和实现相关模型和算法,并基于计算机的运算对输入的图像的图像语义和图像内容进行识别,从而让计算机像人类视觉一样能够明白图像所传递和表达的意思。图像理解研究的应用领域相当广泛,在医学医疗,安全控制,军事科技等领域都能见其身影,但是由于应用需求和应用范围的不断深入和拓宽,图像理解这一研究领域正在受到更加多的关注。本文在总结分析了近年来国内外对于图像理解研究领域的相关研究成果后,首先提出了分层语义模型的概念。分层语义模型通过对于语义空间中所涉及的图像语义的分析,能够将语义空间中的图像语义构建成具有上下层关联的语义模型。论文在提出分层语义模型的同时,还给出了对于图像语义相互联系和自身属性的定义。在提出了分层语义模型的概念基础上,本文继而提出了语义绑定的分层视觉词汇库的概念,并阐述了其构建的方法和讨论了相关细节问题。语义绑定的分层视觉词汇库是在分层语义模型的模板上而建立起来的基于SIFT(Scale-Invariant Feature Transform)图像特征的视觉词汇库,它是由具有分层结构的若干子词汇库组合而成,每一个子词汇库都与一个特定的图像语义相绑定。本文在提出语义绑定的分层视觉词汇库之后会给出其与传统BOVW(Bag Of Visual Words)所产生的视觉词汇库的比较分析。本文最后把分层语义模型和语义绑定的分层视觉词汇库理论应用到两个具体的图像理解问题中去:1)基于语义的图像内容识别问题研究;2)基于内容的图像检索问题研究。本文将会具体阐述通过本文提出的模型算法生成解决上述两类研究问题的解决方案。同时本文还将通过基于上述两类研究问题的仿真实验,以及同传统算法模型性能的比较来充分说明本文提出的模型算法的创新性和有效性。更多还原

【Abstract】 With the development of the technology of internet and multi-media, the application of Digital Image has been fully spread through the social life.And at the meantime, with the continuous progress and innovation of the hardware and software facility, Computer Science is booming on its way. Under such circumstances, the research on Image Understanding has become one of the hottest points in field of Computer Vision. The Image Understanding is mainly responsible for recognizing the semantic and content of images just like what human beings do with the help of the proposed models and algorithm which is running on computers. The theory from the research on Image Understanding has been widely applied into the society, including the field of medical treatment, security control, military technology and etc. In recent years, More light has been shed on the research of Image Understanding since the need and range of its application is widened all the time.This papger firstly proposed the concept of Hierarchical Semantic Model on the conclusive analysis of recent research work on Image Understanding globally.Hierarchical Semantic Model can construct a semantic model in which semantic is connected with other semantic located at contiguous layers through the analysis on all the image semantic from the semantic space. The definition of the semantic connection and semantic attributes will be given out when the Hierarchical Semantic Model has been introduced.After the introduction of Hierarchical Semantic Model, the concept of Semantic Binding Hierarchical Visual Vocabulary (SBHV) will be proposed, and also the method to construct the SBHV and some certain details will be discussed afterwards.SBHV is a kind of visual vocabulary which is coustructed on the template of Hierarchical Semantic Mode with the SIFT (Scale-Invariant Feature Trasform) image feature.SBHV is made up of several layers of sub-vocabulary which is responding to one certain image semantic.After that, the comparative analysis with traditional BOVW model will be generated.This paper will apply the Hierarchical Semantic Model and SBHV into two kinds of concrete Image Understanding problems: 1) image content recongnition based on semantic; 2) image retrieval basec on semantic. The method to merge the proposed model and algorithm into the solutions of the above two kinds of problems will be discussed. And at the mean time, experiments on applying the SBHV into the solutions of above problems and on comparison with the tranditional model and algorithm is carried out to confirm the innovation and effectiveness of model and algorithm proposed by this paper.更多还原