基于Retinex理论的全变差图像分割方法

Retinex-based total variation algorithm for image segmentation

图像中出现的灰度不均匀和伪影等问题严重影响图像分割的效果。文中根据Retinex理论,充分考虑反射(图像结构)部分分片常数的特性,结合全变差(TV)正则项,提出一种新的变分图像分割模型。利用交替极小化方法和算子分裂技巧,给出该模型的快速数值求解算法。数值实验结果表明,该模型在得到精确分割结果的同时,能有效对图像进行偏移场矫正和伪影检测。

Image segmentation results are seriously affected by the problems of intensity inhomogeneity and artifact.According to Retinex theory,the piecewise continuity of the reflection is considered.Combining with the total variation(TV)regularization term a new variational model for image segmentation is proposed.Furthermore,an efficient algorithm for numerically solving the model is given by using the alternating minimization method and a variable-splitting technique.Numerical experiments show that the model can obtain accurate segmentation results,and improve its performance on the bias correction and the artifact detection.