一种灰度直方图与方差修正的图像阈值化算法

A thresholding algorithm with the modification of gray histogram and variance

针对图像的目标与背景分布方差较大时,Otsu算法分割效果不佳的问题,提出一种改进的阈值化算法。基于像素点的灰度概率构造修正函数,作为类间方差的加权因子,建立新的类间方差准则;将修正的类间方差与类内方差构造复合函数,作为新的阈值化准则函数,根据图像直方图信息进行阈值范围优化;通过准则函数在优化阈值范围内选择最优阈值。实验结果表明,对于目标与背景的方差差异较大的图像,改进算法能够获得准确的阈值化结果,并且在正确提取出目标的同时,能够保持良好的目标轮廓完整性。另外,阈值范围优化使得算法执行效率有所提高。

An improved thresholding algorithm is proposed in this paper to deal with the poor segmentation effect by Otsu algorithm when the variance of object and background distribution is large.In this algorithm,a new between-class variance criterion is established by constructing the correction function based on the gray probability of pixel points as the weighted factor of between-class variance.The modified between-class variance and within-class variance are used to construct the composite function as a new threshold criterion function.Then,the threshold range is optimized according to the image histogram information.Finally,the proposed criterion function is used to select the optimal threshold within the optimal threshold range.Experimental results show that for the image with large variance difference between the object and the background,the proposed method can obtain accurate thresholding results and correctly extracts the object while maintaining good image edge contour integrity.In addition,due to the optimization of threshold range,the efficiency of the algorithm is also improved.