摘要： 设计了一种基于声纳纹理信息的由粗到精的分级图像识别系统。系统的粗识别部分是由灰度—空间直方图构成,该直方图将灰度和空间信息有机地融合起来,较好地解决了传统直方图缺少空间信息的不足;精识别部分通过计算图像频谱曲线的分形维数进行图像识别。该方法既考虑了图像的全局特性又包含了细节信息,从而在不增加计算复杂性的前提下,使识别效率明显提高。通过对声纳图像的识别实验,进一步证明了该方法的有效性和工程应用价值。

关键词: 灰度—空间直方图, 分形维数, 声纳图像, 图像识别

Abstract: Based on coarse-to-fine-recognition, a hierarchy sonar image recognition system is designed. For coarse-recognition, gray-spatial histograms, integrating gray and spatial information, are exploited and overcome the limits of spatial information of traditional histograms. For fine-recognition, the Fourier spectrum in polar coordinates is measured by the fractal dimension. This system contains detail information as well as global information of images. The combination of coarse and fine-recognition improves recognizing accuracy without increasing the complexity of computation. The effectiveness and practical value are shown by sonar image recognizing experiments.

中图分类号: 

• TP391.41