一种基于Tsallis相对熵的图像分割阈值选取方法

A Threshold Selection Method for Image Segmentation Based on Tsallis Relative Entropy

在工业实践中,成像环境恶劣且难以控制,导致图像复杂。对复杂成像条件下的图像实施分割并不容易,针对这一问题,结合Tsallis相对熵及高斯分布提出一种新的图像阈值分割方法。该方法运用高斯分布拟合分割后图像直方图分布信息,将Tsallis相对熵做为分割前后图像直方图信息损失的度量工具。在对图像实施分割时,通过在图像灰度级范围内对自定义的准则函数最小化获取最佳分割阈值。最终将该方法与已有方法在工业无损检测及合成孔径雷达图像的分割实验中进行对比。结果表明,该方法获得的结果视觉效果好、分割精度高、误差小而且算法耗时较少,因此具有较好的应用推广前景。

In the field of industrial practice, the images are complicated because the conditions of imaging are usually poor and difficult to control. The image segmentation for complex imaging conditions is not easy. To solve this problem, a new threshold segmentation method is proposed based on Tsallis relative entropy and Gaussian distribution. In the method, the gray level histogram of image after segmentation is fitted by Gaussian distribution, and the difference between the histogram of original image and the fitted histogram is measured by Tsallis relative entropy. The optimal threshold is determined by minimizing the Tsallis relative entropy. Finally, the perSormance of the proposed method is compared with the several methods on segmentation of non-destructive testing images and synthetic aperture radar image. The results demonstrate that the proposed method has better visual effect, higher precision of segmentation, smaller segmentation error and less computational time. Thus, the proposed method has a good prospect in further applications.