A GA-Based Automatic Pore Segmentation Algorithm

Pore feature is important for hardwood identification. But it's difficult to segment pores from wood cross-section images since pore, fiber and longitudinal parenchyma in the image are similar in shapes but different only in size, and the different hardwood species varies in the size of pores. In order to segment pores automatically without parameters set manually, it is necessary to design an adaptive algorithm which may be applied for all kinds of hardwood cross-section images. In the paper, an adaptive method is proposed to evaluate the optimal threshold of closed region area for pore segmentation. The method sorts all closed regions according to the area and classifies closed regions into two classes with maximum between-class variance method. We implements the method based on genetic algorithm to overcome the drawback of being time-consuming. Experiment on images of hardwood species shows that the threshold obtained by the genetic algorithm is very close to but more efficient than the ordinary enumeration algorithm. Moreover, with the obtained threshold majority of pores can be extracted except for some very small ones.