Abstract: A method of computation for multi-window local fractal feature is presented，a novel segmentation algorithm based on the method and traditional image segmentation algorithm is proposed to deal with target under complex environment.Firstly，exponential wavelet at one scale is used to filter image in order to reinforce the edge of image.Secondly，it exacts the fractal intercept of original image using multi-window approach according to the intensity of filtered image，and realizes target detection using region growing algorithm.Finally，morphological operation is utilized to extract target.Experiments show that the proposed method can segment and extract the area of target under complex environment effectively and its low calculation is also proved.

Key words: image segmentation, fractal, multi-window, fractal intercept, region growing algorithm

摘要： 提出了新的图像多窗口局部分形特征计算方法，并结合传统的图像分割方法，给出了一种新的复杂背景下人造目标分割算法。用指数小波滤波器在一个尺度上对图像进行滤波，突出图像的边缘信息。根据滤波后图像像素点的亮度大小，对原始图像进行多窗口几何度量空间变化率的计算，在此基础上，运用区域生长实现目标分割。对分割得到的图像进行数学形态学处理，提取出图像的目标。实验表明，该算法能够有效地实现复杂背景中的目标分割，并且计算复杂度低。

关键词: 图像分割, 分形, 多窗口, 几何度量空间变化率特征, 区域生长