基于形状先验和梯度约束的彩色图像分割

Color image shape prior and the gradient constraint based segmentation

针对图像遮挡、噪声等复杂场景下,仅依赖颜色信息难以准确分割的问题,将形状先验和图像梯度分别引入马尔科夫随机场框架中,提出一种基于形状先验和梯度约束的彩色图像分割方法。该方法基于颜色特征和形状模版定义能量函数,梯度信息的引入允许待分割目标与形状模版间有一定差异,且待分割目标与形状模版间的变换具有仿射不变性,整个能量函数通过图割算法实现能量最小化,得到最终分割结果。实验结果表明,该方法具有有效性。

Considering image is occluded, noisy in complex scenes, image can not be segmented correctly only depend on image color information, shape prior and image gradient are introduced to the markov random field（MRF） framework, a shape prior and image gradient based color image segmentation method is proposed. In the method, the energy function is defined by both image features and the shape template, image gradient information is exploited to permit the slightly difference between the target object and shape template, and the shape alignment procedure allows the affine invariance between the target object and shape template. The whole energy function is minimized through Graph cuts method and the final segmentation results are given. Experiment results show that the proposed method is effective.