Active Contour Models


Active contour models or snakes are a promising and vigorously researched image analysis technique. They offer a unique and powerful approach to image analysis that combines geometry, physics, and approximation theory. A priori knowlege of object shape to constrain the segmentation problem can be used and the inherent continuity and smootheness of the models can compensate for noise, gaps and other irregularities in object boundaries. Finally, active contour models also support interaction mechanisms that allow the expert to influence the segmentation results.

The result of the snake energy minimization process depends on the snake's initial position, internal elasticity parameters and the image forces which attract the snake to salient image features (usually object's edges). Internal elasticity parameters give snake a property of elastic material. The elasticity parameter a gives the snake a property of an elastic, rubber band, which tends to shrink, while rigidity parameter b gives it a property of a steel band, which tends to straighten the corners.

TThe following images show the effect of internal energy in the absence of external energy. The initial position in the shape of a capital letter H is modified in 50 iterations using different values of elasticity parameters.

Figure 1: Parameter b=0, while parameter a takes values of 0.1, 1, 10 and 100 respectevly.

Figure 2: Parameter a=0, while parameter b takes values of 0.1, 1, 10 and 100 respectevly.

The initial position of the snake is the only description of the snake's shape and size. We have developed model driven snakes