3 . Snake or Parametric Active Contour

Many of the original or classical image segmentation techniques like thresholding or edge detection techniques are very susceptible to noise. Another issue that causes problems are gaps in the boundary. As seen in chapter two watershed methods also perform poorly under these conditions. The snake or parametric active contour was developed to handle these problems. They perform well in spite of noise and boundary gaps. This chapter has been done in cooperation with Aurelie Bugeau an exchange student from the 'Higher National Engineering School of Electronics, Computer Science and Telecommunications of Bordeaux' in France, who investigated and implemented certain enhancements on the snake method. First we treat the theory of snakes, then the implementation of the snake is given. Some improvements that have been developed are discussed in the following sections, as applying the snake on multiple scales and with the gradient vector flow force. The balloon snake and a semi-automatic initialization developed by Aurelie Bugeau are discussed next. Then the robustness of the snake method is tested by applying it to some artificial and real images. Finally a discussion on this method is given and some suggestions for future enhancements.