【摘要】 本文用子波变换的方法描述了纹理图像多尺度、多方向的特性,提出了适合于纹理图像分类的新的子波特征。通过对其稳定性和视觉特性的详细分析,指出此特征优于传统的能量特征。文章最后结合九类自然纹理图像,分别基于标准子波特征、子波包特征用BP神经网络进行了分类识别。实验结果表明,在无噪声情况下,对自然纹理图像可无误差分类;在有噪声情况下,正确分类识别率高,表现出强的稳定性。更多还原

【Abstract】 This paper describes the characterization of texture properties at multiple scales and orientations using the wavelet transform, and introduces a new wavelet feature suitable for textured image classification. It is pointed out that the new feature is superior to conventional energy measurement by analyzing its stability and its visual proterty in detail. Finally, nine kinds of natural images are classified successfully based on wavelet feature using BP neural network. The results demonstrate natural textured images can be classified without error and done at higher correct classification rate under white noise.更多还原