【摘要】 区域生长和动态聚类分析相结合,是目前常用的串行图像分割技术。常规区域生长算法仅仅依赖灰度分布聚类的分割效果缺乏鲁棒性,对于光照不均匀的图像更容易失效。提出的方法以区域增长方式为识别分割目标的过程:在灰-梯散射图的特征空间内定义特征差异函数以表现同质性区域的不同特征,并以定义的样本散度作为区域生长的准则函数。在区域增长过程中,象素的归类将沿着样本散度减小的方向,从而有效提高了聚类效果。相比其他的区域生长算法,该算法依据图像的灰度分布特征确定聚类参数,分割效果具有分割效果鲁棒性好、对高斯噪声不敏感的特点。更多还原

【Abstract】 Region-growing combined with dynamic clustering analysis, serial method in common use, always lack the robustness due to that it merely depends on the gray distribution, especially when it deals with the inhomogeneous images. A new region-growing method is based on a characteristic-difference function defined in the characteristic space of 2-D histograms. The minimization of the sample-divergence, which is defined as the role function in growing, determines the classification of the pixels. The realization of the method proves better robustness while less sensitivity to Gauss noise.更多还原