结合Gabor纹理特征的局域化多通道水平集分割方法

摘要

本文提出了一种局域化多通道主动轮廓模型的图像分割算法.针对纹理特征比较明显的图像,通过Gabor滤波提取纹理特征,与图像灰度信息构成多通道.考虑到演化过程中曲线内部和外部特征属性不均匀,引入局域化思想,通过计算各像素在局部区域的最小能量得到图像分割结果.最后算法结合先验形状对有遮挡目标进行分割,并能得到理想结果.大量实验验证了该方法具有良好的分割性能,优于同类算法.%An new algorithm based on localized Multi-Channel active contour model is proposed for image segmentation. For the images with obvious texture, Gabor texture features ate constituted multiple channels of active contour model together with image intensity information. Considering that intensity and texture characteristics are inconsistent in the interior and exterior of the evolution curve, localized energy idea is introduced, the minimum energy is calculated in the specific local area around each pixel on the evolution curve. Our model combined with the shape prior is used to segment the shadowed objects. The proposed algorithm is exemplified on various images objects and its superiority over state of the art variations] segmentation techniques is demonstrated.