Adaptive Smoke Image Segmentation Algorithm Based on Improved Gaussian Mixture Model

With the aim to extract the gray（white） smoke and black smoke simultaneously from the smoke image segmentation, the adaptive smoke image segmentation algorithm based on improved Gaussian mixture model is proposed by using the characteristic of the smoke belonging to foreground object and the color feature of different smoke. On the basis of Gaussian mixture model, this method removes long time no match expired Gauss component to shorten background modeling time. Since the smoke agglomeration features and R, G, B three components relationship of gray（white） smoke and black smoke, the algorithm of smoke suspicious area extraction adopting adaptive differential threshold for different blocks is presented. For the light changes suddenly and the object moves into the scene and still down makes the scene changes gradually, the global and local background renewal is required. Furthermore, a correction factor is introduced to revise updated background image again to ensure that the background image is close to real scene. The difference between current image and background image is used as the base of block difference threshold, and the real-time block difference threshold is updated. Simulation results show that the improved adaptive smoke image segmentation algorithm can extract both gray（white） smoke and black smoke simultaneously, and the target edge irregular information is saved completely. At the same time, the accuracy of proposed algorithm is improved by 35.9% and operating time becomes shorter compared with existing algorithms.