基于多尺度曲面拟合的细胞图像表面分割算法

Algorithm for cell segmentation based on multi-scale surface fitting

针对不规则多形变形状变化特征的红血球细胞分类识别问题,提出一种基于阴影形状恢复技术与多尺度曲面拟合相结合的复杂表面分割方案。利用Shape from Shading进行重建得到的细胞表面三维高度场可用于计算高斯曲率与平均曲率并得到表面类型图像标识的初始分割,然后通过多尺度二元多项式函数分别对不同类型表面进行曲面拟合,得到的分割结果可将不同形状部分区分开来,并以此作为进一步对细胞进行分类的依据。实验结果表明,该方法易于实现,且分割效果较好。

In order to recognize red blood cell automatically, an approach is proposed to segment complex surface based on shape from shading technique and multi-scale surface fitting in regard to its irregularity and deformability. The reconstructed 3D height field helps calculate Gaussian curvature and mean curvature, which can be used to produce surface type label image. This coarse segmentation can be further divided into different regions though multi-scale bivariate polynomials functions fitting. The final segmentation result would be used to classify them in the next step. The experimental results show this approach is easy to implement and have good result.