一种图像纹理特征检索算法

Content-based image retrieval based on Radon and wavelet transform

提出一种基于Radon和小波变换的图像纹理特征检索算法,对原始图像进行坐标系的旋转校正,确定方向归一化的图像,再对方向归一化后的图像进行Radon变换.根据Radon变换投影数据的几何特性,构造了适合投影数据的具有尺度和平移不变性的小波分解,该小波分解系数具有旋转、平移和尺度不变性.采用图像中各尺度小波系数的能量值作为图像的纹理特征,以此进行图像检索.基于纹理特征的试验结果表明,该特征具有旋转、平移和尺度不变性,与RIM算法相比平均检索率提高了3.8%.

A new texture feature is proposed using the Radon and wavelet transform according to the geometric property of Radon transform and a new image retrieval algorithm based on the texture feature is presented. First, the orientation-normalized image is achieved by rotating its principal axis to the horizontal direction. It is turned to obtain the rotation-invariant project after the orientation-normalized image is done by the Radon transform. Second, the adoptive wavelet decomposition, which has the property of translation-and scale-invariance to the project of Radon transform, is constructed. Finally, the energy of every sub frequency band is extracted to be the texture feature of the image. The image retrieval experiments indicate that this method has the property of shift, scale and rotation invariance and has a higher retrieval-rate of 3.8M than the RIM method＇s.