基于玻耳兹曼原理的图像分割算法

Image Segmentation Algorithm Based on Boltzmann Theory

针对图像分割算法中存在的难以自动处理和范化能力差等问题,基于热力学的玻耳原理和Metropo lis准则,设计了一种新颖的图像分割算法.该算法在无需预先设定阈值的情况下,可高速地获得单像素、连续轮廓边界的分割图像,并通过调节温度系数,实现图像由概略到细节的分割,因此避免了过分割现象.该算法的分割时间随温度系数的减小而增加,而温度系数的减小造成分割类数增多所导致的像素点重复计算问题,可通过并行计算来改进.由对比实验证明,该算法在快速性、自动处理程度和范化能力性能指标上,明显优于常用的K聚类和可控竞争惩罚学习等图像分割算法.

Aiming at image segmentation algorithms problems, such as difficult auto processing, weak normalizing, etc., based on the Boltzmann theory and the Metropolis rule, a novel image segmentation algorithm is proposed, where the one-pixel-wide segmented contours remaining continuous can be acquired quickly without setting pre-thresholds. Simultaneously, the segmentation from rough to detail can be realized by adjusting the temperature parameter to avoid the over-segmentation. The segmentation time is increased with the reducing temperature due to the augment of the cluster's number. The problem of the pixel repeated computing can be solved with parallel computation. The experimental results indicate the better performance of efficiency, automatization and normalization of the proposed algorithm than the K-means and rival penalize controlled competitive learning algorithms.