复杂自然背景中人造目标的分割研究

A Image Segmentation Method based on Local Fractal Dimensions and Wavelet Transform

在以自然景物为背景的图像中,分形维数特征是一种能有效将人造物体从自然背景中分割出来的一种纹理特征。提出了一种基于G abor滤波器及差分盒维数的分形纹理特征提取方法,该方法依据不同的滤波器尺度分别采用不同尺寸的滑动窗口来计算分形纹理特征。并利用FCM方法实现了对图像的分割。实验表明该方法能很好地对真实自然背景的图像进行分割,并能由此获得人工目标的轮廓图像。

In the case of image with a nature scene as background,the fractal texture feature is an effective feature that can separate the artificial object from the nature scene. In this paper,a texture image feature extraction method,which based on Multi-channel filter and differential box counting （DBC） fractal dimension estimation method ,is brought out. Here an flexible size of sliding window is adopted according to the scale of Gabor filters when estimate the fractal features of image. At last,the segmentation of image is archived with Fuzzy c-means method. Experiment result shows that this method can successfully segment those images of real nature scene with artificial object.