【摘要】 提出了一种基于快速小波包直方图技术的图像检索新算法。此方法主要有图像的小波包分解,最主要能量频带的选择和小波包直方图的抽取及相似性度量三个步骤。首先,用一族正交小波基分解一幅图像并用小波包系数计算各个频带的能量;其次,选择几个最主要能量频带进行阈值化和非线性滤波;最后,抽取小波包直方图作为特征表示并应用直方图相交距离从图像数据库中检索被查询图像。由于该方法在特征抽取中应用较小的特征空间,因此需要较小的计算复杂性。实验结果表明,这些技术在图像检索中可以获得更好的性能。更多还原

【Abstract】 A novel image retrieval algorithm based on fast wavelet packet histogram techniques is presented. The algorithm consists of three steps: wavelet packet decomposition, most dominant energy channels selection, wavelet packet histogram extraction and similarity measurement. Firstly, an image is decomposed by a family of real orthonnormal wavelet bases and the energy is computed using wavelet packet coefficients. Secondly, few number of most dominant energy channels are selected for thresholding and non-linear filtering. Finally, the wavelet packet histogram is computed as feature signatures and histogram intersection distance is employed to retrieval queried image from image databases. The proposed method employes a smaller feature space and involves as less computation cost in the feature extraction. The experimental results show that these techniques can archive higher performance in the image retrieval.更多还原