【作者】 史婷婷；

【导师】 李岩；

【作者基本信息】 华南师范大学 ， 计算软件与理论， 2007， 硕士

【摘要】 计算机技术、多媒体技术以及Internet技术的长足发展导致大量图像的出现，如何快速有效地从大规模图像数据库中检索出需要的图像是目前一个相当重要而富有挑战性的研究课题。基于内容的图像检索技术的研究正是为了解决利用自动获取的图像特征，从图像数据库中检索出相关图像的问题。近年来，此项技术的研究非常活跃，在许多领域都有应用。然而，尽管基于内容的图像检索技术研究已达数十年，取得了长足的进展，但仍无法满足用户的检索要求。最难以解决的一个问题是：表征图像内容的低层视觉特征与用户检索时具有的高层语义之间存在差距。其实，这也一直是计算机视觉、图像理解和模式识别领域的一个难点问题。研究图像的低层视觉特征，以及图像间相似性的度量既具有重要的意义，也面临巨大的挑战。本论文围绕基于内容的图像检索中的一些关键技术，包括提取图像低层特征、图像间相似性度量以及检索算法的评价准则技术，进行了一些探索性的研究：1．提出一种颜色不变量模型，并将其应用到颜色相关矩阵表达法。同时，进一步利用二值化改进了颜色相关矩阵，有效地抑制背景色影响；2．提出颜色向量角和欧几里得距离共同检测彩色图像边缘，有效地弥补了仅使用颜色向量角或欧几里得距离进行边缘检测的缺点；3．提出新的空间特征表达，弥补了基于内容的图像检索方法在空间特征表达方面的不足。同时，提出sin相似性度量法则，进一步提高检索精确度；4．改进客观图像质量评价领域的最新成果SSIM，提出一种研究CBIR的新思路。在此基础上，研究了国内外的一些图像检索系统，发现检索算法的测试研究和算法性能比较方面存在不足，主要表现在只有先建立自己的图像检索系统，才能进行算法测试，而开发图像检索的实验环境是很复杂的。为了解决上述的问题，设计并实现了通用的基于内容的图像检索试验平台SttImageRetrieval，且高效易用，并通过模拟实验证明算法的可行性。更多还原

【Abstract】 Due to the steady growth of computer, multimedia,and Internet techniques,a hugeamount of images are available.Currently, rapid and effective searching for desiredimages from large-scale image databases becomes an important and challengingresearch problem. CBIR (Content-based image retrieval) is researched to solve theproblem of retrieving relevant images from an image database based on automaticallyderived image features.In recent years, CBIR is a very active research direction andhas been applied to many fields.After several years’ research, CBIR has made the considerable progress,however, it was still unable to satisfy the users’ requests. The most difficult question is:there exists a gap between low-level visual features and high-level semantics.Actually,this also is difficult in the computer vision, the image understanding and a patternrecognition domains.Study on low-level visual features and the similar measurealready have the vital significance, also face the huge challenges.In this paper, lots of exploratory research work has been done around some keytechnologies of CBIR, including low-level feature extraction, similarity measure aswell as algorithm appraisal criterion technology:Firstly, one kind of color invariance model was proposed, and applied it into thecolor co-occurrence matrix.Meanwhile, further the binaryzation improvementsuppressed the background color influence effectively;Secondly, color edge detection using jointly Euclidean distance and vector anglewas proposed, making up shortcomings of only using one of the two effectively;Thirdly, the new spatial expression was proposed, making up insufficiency ofspatial expression aspect in CBIR. Meanwhile, sin similarity measure wasproposed to further improve the precision of retrieval;Fourthly, SSIM(Structural Similarity Index),the latest result of image qualitymeasure was improved, which was one new idea of CBIR research.Based on the study of the domestic and international image retrieval systems, there are some deficiencies regarding the retrieval algorithm test research and thealgorithm performance comparison aspect, mainly manifesting in that our own imageretrieval system must be established at first, algorithm tests can be carried out. Whileestablishing image Retrieval environment is very complex. In order to solve the aboveproblems, a common content-based image retrieval test platform calledSttlmageRetrieval has been desiged, which is user-friendly and efficient, and thispaper proved the feasibility of algorithms through simulation experiments.更多还原