Shape Modeling is a key step in many computer vision applications. Identifying shapes provides meaning to the image which is otherwise just a collection of intensity values. Deriving what meaning these shapes convey is the typical task addressed by recognition. A simple approach to shape detection, like thresholding, may not result in very meaningful segments in the image. Active contours solve this problem by imposing constraints on the contour of the shapes obtained ,like smoothness and continuity. The boundaries evolved under these constraints are more likely to correspond to a real world object than those obtained from a simple thresholding of the intensity values.
The other use of shape detection is to provide a compact description of the image. This compact description can reduce the complexity of the algorithms that are used to process the image later. It can also be used to achieve image compression.