基于直方图互确认的图像阈值化分割

Image thresholding based on mutual recognition of histogram

目前图像阈值化中存在的主要问题是分割结果的有效性以及分割数目的确定,本文提出一种新颖并且原理简单的图像阈值化算法.根据图像在直方图上的直观分布,提出一种新的阈值判断机制,该机制基于峰谷对阈值的影响,累积所有直方图对阈值的认可程度,并结合一些加速策略,使得算法具有很强的实用性和适用性.算法结合最终的认可程度的百分比确定取值数目,进而确定分割数目.在实验中,引入两种较新的阈值化方法,本文算法的分割效果在实验比较中表现良好,也很稳定.该方法在红外图像处理中也得到较为显著的分割效果.

For more effective image thresholding,a novel and simple method was proposed which based directly on the histogram of the image.According to the intuitive appearance of the histogram and the influence of peaks and valleys to final threshold,a new thresholding measurement is defined creatively.Accumulating mutual recognitions of all other histogram bins and combining some speeding-up strategies,the proposed measurement makes image thresholding more practicable and applicable.The number of segmentation is determined by the percentage of the final recognition value.As can be seen from the comparison with other methods in experiments,results of the proposed method looks better than other ones.Moreover,the objects are highlighted well in infrared image.