视觉注意特征的变分水平集图像分割模型

针对传统主动轮廓模型较低的鲁棒性能和对先验知识融合能力的不足,基于视觉注意机制的先验知识和曲线演化的理论框架,首先建立图像底层视觉显著性特征的数学模型,在此基础上提出新的曲线演化能量泛函模型,然后对该能量泛函采用变分水平集方法进行推导,得到曲线演化的偏微分方程,数值实验表明该模型相对于经典主动轮廓模型具有更强的抗噪性与分割效率.该模型的提出为进一步在主动轮廓模型中引入更高层次视觉显著性特征、得到更优越的分割模型打下了基础.

Model of the variational level set image segmentation based on visual attention features

The robust and fusion capacity of the traditional active contour models is poor.The mathematical model of rock-bottom visual attention characteristics in image was first established based on a priori knowledge of mechanism of visual attention and theoretical framework of curve evolution,a new curve evolution energy functional model was put forward,then partial differential equations to guide the curve evolution were established according to variational level set to this energy functional.The numerical experiments showed that the model was more robust and had higher segmentation efficiency than classical active contour model.The model laid the foundation for higher level visual significant features and getting better segmentation.