An adaptive multi-seed geometric active contour model for river recognition

This paper presents a novel multi-seed vector-valued framework for river recognition based on geometric active contours. There are four core components of this framework: vector-valued scanning algorithm, a geometric active contours model, low resolution segmentation with elevation data and high resolution optimization.The combined algorithm allows for a rapid evolution of the contour and a convergence to its final configuration with a small number of iterations. Compared with the conventional segmentation methods, our approach has the advantages of dealing effectively with complicated satellite images, automatically initializing a number of snakes based on color and texture features, accurately and rapidly identifying the target objects with elevation data.