Color Smoke Image Recognition Based On Fractal Dimension

In recent years, a large number of forest fire accidents caused incalculable loss to the country and the people. Therefore, to propose an effective fire detection method has always been extensively researched by the scholars. Traditional fire detection technology uses the flame temperature, smoke concentration and illumination intensity to detect whether the fire broke out. Currently, as an important means of preventive fire, the image-type smoke detector has been hot topic, also it can prevent fire and take emergency measures to avoid fire.This paper presents a new smoke identification method combining the fractal theory and color method. By using the smoke image texture information, the image is divided into the smoke area and background. According to the peculiar color characteristics of smoke image, if the divided region has met the color characteristics of the smoke, in can identify smoke. The specific process is as follows:(1) Gray image segmentation based on fractal theory. Aiming at the different fractal texture features between the forest and smoke image, a segmentation algorithm method combining the fractal and Otsu algorithm is proposed. On figuring the fractal dimension, we respectively improve algorithm from fitting error precision and speed of algorithm. By calculating the gray level in sub-window, error of fitting accompanied by figuring the fractal dimension decreases and dose better expression of surface texture fractal image; By extending the range of sub-window size, the number of boxes covering the sub-window decreases, and so does the execution time of the algorithm. The simulation results prove that the modified differential box-counting dimension method can better express the image surface texture Information, and combined with Otsu method can segment smoke region.(2) Smoke recognition based on color smoke image. The colour image has rich color information, studying the color information is an important method to identify smoke. According to the characteristics of the smoke in the RGB and HSV color space, we can extract the reference points in RGB color image,also, we can cluster the pixels whose color is similar with the reference points based Human Visual System. Then the smoke recognition method based on color feature can more accurately identify the smoke. Using the above method can verify the color information in the fractal segmented region, it dose match the characteristics of smoke. Experimental results show that this paper presented methods is very effective.Therefore, smoke image recognition combined fractal theory with color information not only can be used as a direction of the study of smoke recognition but also provides a good way to prevent fires.