基于颜色与纹理融合的图像特征提取与检索方法研究-信号与信息处理专业论文.docxVIP

摘要 摘要 I万方数据 I 万方数据 摘 要 本文针对现今各个领域中日益增长的图像信息检索应用需求，在系统的分析了国 内外基于内容的图像检索技术的发展现状和所面临的问题的基础上，对颜色、纹理特 征提取算法以及相关反馈算法进行了进一步探索。主要创新点如下： （1）针对全局颜色特征无法捕捉图像空间信息的缺陷，以及颜色直方图算法中 处于量化临界边缘的颜色所面临的归属问题，本文将图像分块与提升小波算法相结 合，提出了分块提升小波算法来提取颜色特征。 （2）针对现有的纹理特征提取算法的不足，本文采用提升小波算法进行纹理特 征提取，提高了纹理描述的准确性。 （3）为了进一步提高检索性能，本文还对图像检索中的相关反馈技术进行了深 入研究，提出了融入交互式遗传算法的相关反馈算法。该算法综合运用正、负反馈图 像信息自动调整特征分量权重、区域权重、颜色纹理比重，并与记忆性和禁忌有哪些信誉好的足球投注网站相 结合，运用遗传算法不断的更新优化示例特征向量，使系统能在短时间内动态捕捉用 户意图。 根据上述算法，本文设计开发出一个基于内容的图像检索试验系统，并进行了算 法有效性的验证，实验结果表明，该系统具有明显的优越性和通用性。 最后，对全文进行了总结，概况了主要研究成果，并对今后的研究方向进行了规 划。 关键词：基于内容的图像检索；颜色特征；纹理特征；提升小波；交互式相关反馈； 遗传算法 Abst Abstract II万方数据 II 万方数据 Abstract Aimed at the increasing application demands of image information retrieval in each field, the algorithms in color, texture feature, and relevance feedback are explored further based on analyzing the present technique about Content-based Image Retrieval and the confronting problem. The main innovations in the dissertation are: Aimed at the defect that global color feature can’t catch the space information of the image, and the problem that the color at the quantification critical edge which quantification area does it should belong to in the algorithm of color histograms, the part lifting scheme algorithm is proposed, which combines lifting scheme and sub regions together. Aimed at the failure of the present texture feature extraction algorithms, the lifting scheme algorithm is used to extract texture feature. The accuracy of texture description is improved. In order to improve the extracting performance, the relevance feedback algorithm is researched in depth. A feedback algorithm blending relevance Genetic Algorithm is proposed. This algorithm uses positive and negative feedback images’ information to adjust feature vector weight, region weight and color-texture weight, and combines with memory and taboo search algorithm, applies the Genetic Algorithm to renew and optimize the example feature vector continuously to catch user’s purposes in a short time. Acco