中文摘要：

为了解决图像在转载过程中所产生的旋转变化和尺 度变化对检索的影响,根据熵的对称性,提出了基于NSCT及熵的旋转不变图像检索算法。首 先,利用非下采样轮廓波变换（NSCT）对图像进行多尺度、多方向分解,对不同尺度、同方 向的高频方向子带求多尺度积,以减小尺度变 化和噪声对检索效率的影响；然后,考虑到图像旋转后各方向子带在整幅图像中的能量比例 不会发生变化, 将各方向子带的能量比例作为概率矢量,各方向子带的粗糙度作为权值求取图像的加权信息 熵,作为具 有旋转不变性的图像纹理特征,利用矩提取图像的颜色和形状特征；最后,归一化3种特征 来比较两幅图 像的相似性。性能测试表明,本文所提出的方法对旋转变换鲁棒性强,且具有很高的查准率和查全率。

英文摘要：

With the rapid development of multimed ia and network,there have been much interest and a large number of researches on digital image.Searching image from massive network images becomes an urgent tas k.In order to reduce the effect of rotation and scale transform in the process of ret rieving,a rotation invariant retrieval algorithm based on entropy and non-subsampled contourlet transform （NSCT） is pr oposed according to the symmetry of entropy.Firstly,the image is decomposed in multi-scale and multi-direction by NSCT,and high frequency sub-bands of the same direction at different scales are multiplied to reduce the effects of scale change and noise. Secondly,because the energy proportion of each directional sub-band in the who le image is constant after image rotation,the energy proportion as probability vector and the roughness of each d irectional sub-band as weight are employed to calculate the weighted information entropy of the image,which is con sidered as the rotation-invariant texture feature of image.Color and shape features are extracted by moments.Finally,three kinds of features are normalized to analyze the similarity between two images using Euclidean distan ce.Rig orous performance tests on two databases of rotation and scale change show that the proposed algorithm is robust to rotation and scale variance,and has high precision and recall.