Automatic segmentation on multiple starch granules from microscopic images

Starch is the most important carbohydrate in the human diet and contained in many staple foods such as rice, wheat, corn, potatoes and cassava, etc. Currently, microscopic imaging technology is one of the most effective ways to investigate and analyze the structure of starches. Segmentation of starch granules is a necessary step for starch granule structure analysis such as starch granules detection, shape recognition, and size computation. This article investigates a new method based on GVF (gradient vector flow) snake deformable model for starch granules segmentation. The proposed method focuses on full automatic segmentation on granules, especially on separation of adjacent and contacted starch granules, which occur widely in microscopic images. A novel energy function based on position and intensity is introduced into the directional gradient computation, thus the directional gradient is used to obtain the directional GVF snake, which drives the deforming contours to the real contours of multiple granules. To demonstrate the good ability of the proposed method, we segment 30 starch granule images and compare it with the level set method, experimental results show that the new method can separate multiple starch granules successfully; especially it works much better on overlapping objects segmentation than the level set method.