基于多尺度Radon变换的图像检索

根据Radon变换的统计特性构造了不变量,提出一种新的基于多尺度Radon变换的图像形状检索方法.对检索图像作小波变换,根据小波模极大原理得到边缘图像,对边缘图像构造Radon变换中心矩,在中心矩的基础上根据Radon的统计原理构造出尺度不变矩.由于矩阵的奇异值具有旋转不变性,因此针对不变矩向量矩阵求奇异值,该奇异值特征向量具有平移、尺度和旋转不变性.将该Radon变换的不变量作为形状特征,并进行高斯归一化,按照欧氏距离计算不同图像间的形状相似度.试验结果表明,该方法对高斯噪音具有较强的鲁棒性,与其它方法相比具有较好的检索效果.

Content-Based Image Retrieval Based on the Multi-Scale Radon Transform

The invariant using statistical theory of Radon transform is constructed and a new kind of image retrieval algorithm based on multi-scale Radon transform is presented.In the first place,the shift invariant is achieved by the central moment on the Radon transform of edge images,which is gained by the wavelet modulus maximum.In the second place,the scaling invariant based on the central moment is deduced after analyzing the statistical characteristics of Radon transform.Since the singular values of matrix are uncorrelated with the position of the column or the row of matrix.Here,this property can be used to get the rotational invariant from the Radon moments.In the third place,the rotational invariant property of singular values on the moment matrix is used to get the rotational invariant.In this way,the character vector with shift,scaling and rotational invariant is constructed in this algorithm.Finally,the Gaussian model is used to normalize the different sub-characters distance to the shape feature of image.The shape similarity of the querying image and other images is computed by the Euclidean distance.Experiments indicate that this method is of robustness to the noises in image′s similarity retrieval and has a higher retrieval-rate than those of Pseudo-Zernike moments,wavelet modulus maximum and Tchebichef moments.