Active contour models for infrared cloudy shapes segmentation
C. Collet, Pierre Thourel · 2002
By nature, the clouds are deformable objects with fuzzy, blurred and indefinite contours. The detection methods usually used (convolution mask for edge detection, linear filtering...) are ineffective for this kind of pattern. Nevertheless, the description of clouds, as closed geometrical patterns, is of great practical interest : it allows one to describe the moving cloudy shape with a lot of geometrical parameters (perimeter, center of gravity, area, energy, orientation of the pattern...) that are not valid for an unclosed shape. Tracking methods for deformable shapes are one of the main lines of research in image processing. This kind of problem is well illustrated by cloudy shapes. The use of an active contour model, called "snake", introduces the concept of deformable closed edges. This article will give an illustration of this new technique for cloudy edge detection and proposes a normalization of energy functions governing the snake evolution. This normalization will allow the final energy level of the snake to be compared for different patterns.>