基于双向多项式拟合的的动态阈值分割算法

A dynamic threshold segmentation method based on bidirectional polynomial fitting

针对角膜内皮细胞图像非均匀光照的问题进行了研究,采用限制对比度自适应直方图均衡化、同态滤波方法进行图像增强^([1])。同时提出了一种基于双向多项式拟合的的动态阈值分割算法,分别对图像矩阵的行向量和列向量进行了多项式拟合得到两个方向上拟合后的阈值矩阵,然后将每个矩阵点的阈值同图像矩阵在该点的值进行比较得出二值分割图像,该方法保留了因光照不均而容易丢失的特征形态细节。最后通过图像孔洞填充、图像腐蚀膨胀、图像细化等一系列的数学形态学运算得到了角膜内皮细胞图像的特征形态图形。实验结果表明,与OTSU法^([2])、Ni Black法等几种算法相比,该算法的分割效果及实时性都具有明显的提升。

This paper researched the uneven illumination problem of the corneal endothelial cell images, using contrast limit adaptive histogram equalization（ CLAHE）, homomorphic filter to enhance image. Meanwhile the paper proposed a dynamic threshold seg-mentation method based on bidirectional Polynomial fitting, each row and column vectors of image matrix were fitted in both direc-tion get threshold matrix. Then the value of each point for threshold matrix obtained were compared with the value of each point for image matrix to obtain final binary segmented image, the method retained details which uneven illumination image easily lost.Finally mathematical morphology operation is used such as image hole fill, image erosion and dilation, image thinning to get the fea-ture image of corneal endothelial cell. Experimental results demonstrate that segmentation results and real- time of this method have sig- nificantly improved, compared with OTSU method, Ni Black method and several algorithms.