Unsupervised active contour model for biological image segmentation and analysis

Active contour models, or snakes, are widely used for image segmentation, especially in the case of biological images. This is because the importance of biological image analysis demands quality segmentation which the active contour models can provide. However, active contours suffer from the serious problem of initialization and tend to wander off toward other image features if initialized far from the actual object. Since a priori information about the regions of interest is generally not available in biological images, we propose to use the conventional hyperstack based multiresolution image segmentation technique to extract information about the regions of interest. This helps in initializing the contours close to the actual object boundaries. The active contour model in the subsequent stage refines the initial contours. Experimental results demonstrate the effectiveness of the proposed scheme in terms of segmentation quality.