Abstract: Aiming at the problem of incomplete flame image segmentation under low contrast, the threshold adaptive segmentation method for color flame image enhancement and the improvement of ant colony algorithm is proposed. First, the multi-scale Retinex enhancement with color recovery is designed for RGB images captured by the acquisition device. The brightness of the background pixel is raised through the brightness control factor. Adding the high-frequency information to the original image from three different scales dynamically improves the local visibility of the flame, thus making the color, texture, and edge information of the flame more prominent. Then, to avoid the randomness of ants during path selection, and enhance ant search efficiency and excellence search, based on the maximum class variance method, the segmentation threshold is automatically obtained from single images by improving the initial pheromone concentration and update rules and inspiration function in the ant colony algorithm, greatly improving the segmentation accuracy and speed. Finally, the proposed algorithm is compared with similar algorithms under different strong light, different scenes and different interference. The experimental results show that the average segmentation accuracy has improved by nearly 1.96 times while ensuring the segmentation speed.

Key words: fire protection, multi-scale Retinex, image enhancement, ant colony algorithm, maximum interclass variance method, flame segmentation