A New Method of Flame Image Segmentation

摘要

Separating flame image from background is a precondition and necessary link for identification for recognition algorithm. In current application, the Otsu method is a simple, effective and threshold value selection method suitable for real time processing, but the limitation of this method is it depends too much on the image pattern of "double peak histogram". To overcome the weakness and shortcoming of Otsu method, this paper puts forward a new method based on grayscale integral projection, for flame image segmentation and positioning. Experiments have proved that this method not only can separate a flame image from background correctly, but also can position the center of a flame image quickly so as to provide the important criterion for dynamic recognition of flame image.