基于布谷鸟搜索算法的图像检索系统设计

发布时间：2018-12-08 21:21

【摘要】：随着互联网的迅猛发展,海量的数据信息与人们的生活紧密相关,图片、视频等多媒体信息迅速增加。如何从海量的信息库中准确、高效的搜索出所需的信息是信息化时代的热点问题。传统的搜索以文字为搜索对象,通过关键字、关键词来实现信息搜索,基于文字的搜索技术已经非常成熟。然而文字搜索的缺陷在于,无法搜索一些很难用文字描述的图片信息,并且文字很难直观全面的表达人们的搜索意图。基于内容的图像检索(Content Based Image Retrieval,CBIR)技术就能够很好的解决这个问题。基于内容的图像检索技术通过上传图片来代替文字搜索,计算机自动提取图像的特征,然后从图像库中找出特征相似的图像。目前,基于内容的图像检索技术需要改进的主要问题是提升搜索效率和减小“语义鸿沟”以提升搜索准确率。本文以基于内容的图像检索为基础做出了以下几方面的工作:(1)提取图像特征构建图像特征库,建立基于内容的图像检索系统。本文以corel1000为图像库,提取了图像的颜色矩、颜色相关图特征以及LBP纹理特征,组成特征向量库,并采用MATLAB为工具,建立基于内容的图像检索系统,实现了通过上传图片来搜索相关图片的功能。(2)提出一种基于内容和布谷鸟算法的图像检索算法,将连续空间寻优的布谷鸟搜索算法应用于离散的图像特征空间进行图像搜索,提高了CBIR系统的搜索效率。布谷鸟搜索算法(CuckooSearch,CS),也叫杜鹃搜索,是由剑桥大学YANG等在2009年提出的一种群智能优化算法,该算法参数少、搜索路径较好、有较强的全局搜索能力。本文将CS算法应用到基于内容图像检索系统中,将图像搜索问题看成寻找最优解问题,利用CS算法搜索路径较好、有较强的全局搜索能力的优点在图像特征空间寻优,最后通过实验证明了该算法比遍历搜索算法在基于图像检索系统中有更高的搜索效率。(3)提出一种基于布谷鸟搜索动态调整支持向量机参数的相关反馈算法,减小了基于内容的图像检索系统中的“语义鸿沟”。首先,将相关反馈问题当作二分类问题,采用支持向量机(Support Vector Machine,SVM)通过反馈结果对图像进行二分类,并通过CS算法动态搜索最佳SVM参数,根据每次反馈结果自适应调整支持向量机参数。通过实验证明该算法比传统的布谷鸟搜索算法、粒子群算法(Particle Swarm Optimization,PSO)以及遗传算法(Genetic Algorithm,GA)让支持向量机更快更准确的实现分类,从而使得图像检索的相关反馈能在更少的反馈次数下得到更高的准确率,提高了搜索准确率。
[Abstract]:With the rapid development of the Internet, massive data and information are closely related to people's lives, pictures, video and other multimedia information is increasing rapidly. How to search the needed information accurately and efficiently from the massive information database is a hot issue in the information age. The traditional search takes the text as the search object, realizes the information search through the keyword, the text based search technology has been very mature. However, the defect of text search is that it is impossible to search for some image information that is difficult to describe in words, and it is difficult for text to express people's search intention directly and comprehensively. Content-Based Image Retrieval (Content Based Image Retrieval,CBIR) technology can solve this problem well. Content-Based Image Retrieval (CBIR) replaces text search by uploading images. The computer automatically extracts the features of the images and then finds the images with similar features from the image database. At present, the main problems that need to be improved in content-based image retrieval technology are to improve the search efficiency and reduce the "semantic gap" in order to improve the search accuracy. In this paper, based on content-based image retrieval, the following works have been done: (1) extracting image features to construct image signature database and establishing content-based image retrieval system. In this paper, the color moment, color correlation image feature and LBP texture feature of the image are extracted by using corel1000 as the image database, and the feature vector library is formed, and the content-based image retrieval system is established by using MATLAB as the tool. The function of searching related images by uploading pictures is realized. (2) an image retrieval algorithm based on content and cuckoo algorithm is proposed. The Cuckoo search algorithm with continuous space optimization is applied to the discrete image feature space for image search, which improves the search efficiency of CBIR system. Cuckoo search algorithm (CuckooSearch,CS), also called cuckoo search, is a population intelligent optimization algorithm proposed by YANG of Cambridge University in 2009. The algorithm has few parameters, good search path and strong global search ability. In this paper, the CS algorithm is applied to the content-based image retrieval system, and the image search problem is regarded as the problem of finding the optimal solution. The advantage of the CS algorithm is that the search path is better and the global search ability is stronger. Finally, experiments show that the algorithm is more efficient than the traversal search algorithm in image retrieval system. (3) A correlation feedback algorithm based on cuckoo search to dynamically adjust support vector machine parameters is proposed. The semantic gap in content-based image retrieval system is reduced. Firstly, the correlation feedback problem is regarded as a two-classification problem. The support vector machine (Support Vector Machine,SVM) is used to classify the images by feedback results, and the optimal SVM parameters are dynamically searched by the CS algorithm. The parameters of support vector machine are adaptively adjusted according to the result of each feedback. Experimental results show that the proposed algorithm makes the classification faster and more accurate than the traditional Cuckoo search algorithm, particle swarm optimization (Particle Swarm Optimization,PSO) and genetic algorithm (Genetic Algorithm,GA). Thus, the correlation feedback of image retrieval can get higher accuracy with less feedback, and improve the search accuracy.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP18

【参考文献】