Saddle point detection for connecting objects in 2D images based on mathematic programming restraints

Saddle points formed through morphologic erosion and existing between adjacent connecting objects in 2D images have been applied for segmenting purposes. In this article, a new approach is presented for searching the saddle points in 2D images using mathematic programming restraints for the purpose of ultimately separating the connecting objects. By combining the pixel distribution information in 3D topographic image and mathematic programming restraints for saddle point, the saddle points in the image can thus be identified. In addition, the relation between step selection in the algorithm and detection rate is also explored. The experiment results on the given real particle images suggest the better robustness in saddle point detection algorithm, which undoubtedly lays the practical and theoretic base for touching object segmentation for 2D images.