【作者】 王令；

【导师】 吴锡生；

【作者基本信息】 江南大学 ， 计算机应用技术， 2008， 硕士

【摘要】 在传统的基于内容图像检索的方法中,由于图像的领域较宽,图像的低级视觉特征和高级概念之间存在较大的语义间隔,检索的效果不很理想。文章研究了图像增强技术在贝叶斯框架下基于内容的感知编组规则的图像检索。经过图像增强技术处理后图像灰暗度及其色彩明暗提高,又通过感知编组提取图像颜色特征进行贝叶斯分类,并根据Lx a x b x空间彩色的距离判定条件来进行检索。经实验验证,该方法的检索效果比通常的方法有较大提高。此外,通常直接采集原始格式的图像检索比较多,由于数据量大,给存储或传输带来不便。文中给出了基于小波变换和二值模式的图像检索方法,其优点在于一方面解决了数据量大、省略解压缩环节、特征向量包含在压缩域检索系数中;另一方面二值模式的图像更有利于提取图像的纹理和形状特征。实验结果表明两者结合提高了检索效率。同时,传统的K均值算法存在两个固有的缺点:(1)对于随机的初始值选取可能会导致不同的聚类结果,甚至存在着无解的情况;(2)该算法是基于梯度下降的算法,因此不可避免地常常陷入局部极优。这两大缺陷大大限制了它的应用范围。而基于粒子群的k-means聚类算法是在传统的聚类算法中引入了粒子群算法。理论分析和数据实验结果表明,该聚类算法克服了传统聚类算法存在的问题,全局寻优能力优于现有的基于遗传算法的k-means聚类算法,且有较快的收敛速度。最后,文中给出了六种颜色空间(HSV、YUV、RGB、XYZ、HSL、YIQ)以及分块加权HSV颜色直方图比较的实验结果,并且综合图像的颜色和纹理特征以及纹理和形状特征进行图像检索时,采用了基于灰度共生矩阵的纹理特征提取和基于不变矩的形状特征提取方法,并在此基础上加入了基于权重调整的相关反馈机制,使用户可以参与检索过程,通过调整权重使得检索结果最终满足用户的检索要求。实验表明,这种算法获得了较好的检索结果。更多还原

【Abstract】 In the traditional approach of content-based image retrieval, the wide image domain results in the wide semantic gap between the low-level features and the high-level concepts. So image enhancement technology is used Bayesian framework content-based sensation grouping image retrieval: After image enhancement technology processing image gloomy degree and its the color light and shade enhance after image enhancement technology processing, and draws the image color characteristic through sensation grouping to carry on Bayesian classification, and carries on the retrieval according to Lx a x b x space colored range estimation condition. Confirms after the experiment, this retrieval effect must be better than the ordinary method. Another, directly collection of the original format of image retrieval is more. As the volume of data, storage or transmission brings inconvenience . Virtues of based on wavelet transform and binary mode of image retrieval lie in solving a large volume of data, omission decompress link, eigenvector included in the compressed domain search foreign coefficient;on the other hand binary mode of image is more in favor of the extraction of image texture and shape features. Experimental results showed that combining the two methods can improve the retrieval efficiency. At the same time, there are two inherent drawbacks in the traditional K-means algorithm:(1) For the initial random selection it may lead to different clustering results, even without the presence of the situation;(2) The algorithm is based on the gradient descent algorithm, it is inevitable that a partial often excellent. These two deficiencies greatly limit the scope of its application.Based on PSO of K-means clustering algorithm is in the traditional clustering algorithm introduced in the PSO algorithm. Theoretical analysis and experimental results show that the clustering algorithm to overcome the traditional clustering algorithm existing problems, global optimization capability is superior to the existing genetic algorithm based on the k-means clustering algorithm, and has a faster convergence rate.Finally,we have compared the experimental results with histogram in six color space and blocking weighted HSV color histogram, gray concurrence matrix and invariant moments were developed to extract texture and shape features in image retrieva1.With the relevance feedback mechanism,users could not only control the retrieval procedure,but also obtained satisfying retrieval results by adjusting relevance weights.The experiment shows that the proposed algorithm is effective and can achieve better results.更多还原