A fire color mapping-based segmentation: Fire pixel segmentation approach

Abstract

Fire detection is a very important task to save human lives and ecological systems. On literature, several fire detection methods use a color mapping function to help on detection process. In this context, we propose a new method based on fire probabilistic color mapping. Using Entropy rules was possible to improve the metric rates. We also numerically evaluate the quality of published fire segmentation techniques and the new one using some segmentation metrics onto two datasets: one for training and test with 226 images and second with 110 images for test. With better performance than compared methods in True Positive (81.33%), Accuracy (89.90%), F-Measure (82.58%) and True Negative rates for not-fire images (98.16%), the results show that our proposed method is more accurate for extracting fire region, indicating the effectiveness contribution of our fire probabilistic color mapping using entropy.