基于视觉侧抑制机理的强鲁棒性图像分割方法  

A strongly robust algorithm of image segmentation based on visual lateral inhibition

作　　者：周理[1] 高山[1] 毕笃彦[1] 何林远[1] 孙毅 

机构地区：[1]空军工程大学工程学院,陕西西安710038 [2]胶带股份有限公司军事代表室,上海200235

出　　处：《中南大学学报（自然科学版）》2013年第5期1910-1917,共8页Journal of Central South University:Science and Technology

基　　金：国防科技重点实验室基金资助项目(9140c610301080c6106);国家自然科学基金资助项目(61175029);航空科学基金资助项目(20101996009)

摘　　要：针对光照强度和图像退化的算法鲁棒性问题,提出一种基于视觉侧抑制机理的强鲁棒性图像分割新算法。该算法充分考虑到人眼视觉侧抑制机理"加强中心抑制周围"的特点,在范围一定的抑制野内,采用双高斯强度系数分布的减法非循环侧抑制网络模型描述神经元兴奋的变化量,与交叉视觉皮质模型相结合形成侧抑制-交叉视觉皮质模型。在此改进模型基础上,利用灰度级-邻域平均灰度级二维直方图将一维交叉熵推广至二维,进而构造出二维最小交叉熵分割判决准则以寻求最优阈值。研究结果表明:针对受噪图像,新算法的分割准则充分利用二维交叉熵增加了图像的局部空间信息,从而取得良好的抗噪结果;针对光照不均和模糊退化的图像,改进分割模型加入视觉侧抑制网络,使得新算法具有较强对比度和亮度适应性以及模糊补偿能力,能够获得精确的分割效果。Aiming at the robustness of illumination intensity and image degradation, a strongly robust algorithm of image segmentation based on visual lateral inhibition was introduced. Fully considering the features of the visual lateral inhibition that the center was enhanced and the surrounding was inhibited, the excited variation of nerve cell was described by the lateral inhibition model with double Gaussian intensity distribution in the inhibition region. Thus, the lateral inhibition- intersecting cortical model was built up. With this modified model, gray level-average gray level histogram was used to turn the 1-D cross entropy into the 2-D. Then, the segmentation criterion of 2-D cross-entropy minimization was formed to seek the optimal threshold. The results show that to improve the noised images, 2-D cross entropy is fully utilized to increase local information of images, resulting in great anti-noise images. For illumination intensity and image degradation, the adaptability of contrast and light together with fuzzy compensation can be obtained from the improved model with the lateral inhibition net, and the accurate segmentation images are also obtained.

关 键 词：交叉视觉皮质模型 视觉侧抑制模型 交叉熵 灰度级-邻域平均灰度级二维直方图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]