摘要
图像分割是图像目标识别和提取的重要步骤。然而由于SAR影像相干成像特点,SAR影像往往信噪比不高,于是传统的图像处理方法难以适用。本文在分析当前主流的阈值法图像分割的基础上,提出一种改进的混合阈值法:首先对影像按一维最大类间方差法求解出一维最佳阈值,以此计算图像分割水体和背景的权重比,然后根据求得的权重比,修改二维最大熵算法,最后采纳遗传算法提高搜索效率,进而求解二维最优阈值实现SAR影像水体提取。为了验证此方法的有效性,针对阳澄湖地区Envisat ASAR影像中同时存在信噪比高的区域和信噪比低的区域设计了仿真实验。实验结果表明,改进方法能较好地从信噪比低的影像中提取水体,同时保持良好的时间复杂度。
Image segmentation is fundamental to target recognition and extraction. However, due to the coherence of SAR imaging, there is a lot of speckle noise in SAR imagery. Therefore, traditional image processing methods are hard to apply. With the analysis of current methods for image segmentation, an improved method based on hybrid thresholding methods has been proposed. Firstly,between class variance （also called Otsu＇s method） is used to compute one dimensional optimal threshold. Subsequently,the weight coefficients between the target and background in SAR imagery are obtained according to the one dimensional optimal threshold. Then the two dimensional maximum entropy method is improved by introducing of the weight coefficients. Fi-nally, genetic algorithm is adopted to improve searching efficiency of two dimensional optimal thresholds. To validate the effi-ciency of this method,simulation experiments are designed for the Envisat ASAR imagery in Yangcheng Lake with the coexis-tence of high signal-to-noise ratio （SNR） area and low SNR area. The results show that the improved method proposed in this paper has good appearance in solving the water extraction of low SNR SAR imagery and keeping proper time complexity.
出处
《遥感信息》
CSCD
2013年第2期62-65,69,共5页
Remote Sensing Information
基金
国家科技支撑计划项目(51079099)
