基于S分量指数加权H分量的圆形直方图阈值法

HSI颜色空间可以用三维空间下的圆锥模型进行解释，基于HSI颜色空间的H分量圆形直方图实现彩色图像分割具有可行性。针对HSI颜色空间的H分量圆形直方图毛刺较多以及相关阈值选取准则分割精度较低的问题，给出S分量指数加权H分量的直方图公式，利用S分量信息对H分量直方图的毛刺进行平滑处理，并通过分析给出指数加权因子的最优取值。在此基础上，提出一种圆形直方图阈值分割法，通过对整个圆形直方图进行整体角均值计算而得出新的阈值选取准则。实验结果表明，该圆形直方图阈值分割法具有有效性，在测试数据集上，与3个圆形最大熵阈值分割法以及2个阈值分割准则相比，所提方法的像素精度值平均提高3.2%，结构相似度值平均提高2.5%。

Circular Histogram Thresholding Method Based on S-component Exponential Weighted H-component

Hue-Saturation-Intensity(HSI) color space can be explained by cone model in three-dimensional space.It is feasible to realize color image segmentation based on the H-component circular histogram of HSI color space.Aimed at the problems of more burrs in the H-component circular histogram in the HSI color space and the low segmentation accuracy of the relevant threshold selection criteria, the histogram formula of the S-component exponential weighted H-component is presented.The burrs of the H-component histogram are smoothed by using S-component information, and the optimal value of the exponential weighting factor is determined.On this basis, a threshold segmentation method of the circular histogram is proposed, and a new threshold selection criterion is obtained by calculating the overall angular mean of the circular histogram.The experimental results show that the circular histogram threshold segmentation method is effective.On the test data set, compared with the three circular maximum entropy threshold segmentation methods and the two threshold segmentation criteria, the proposed method improves pixel accuracy and structural similarity on average by 3.2% and 2.5%, respectively.