Abstract:

The smoke recognition method was proposed which used the differential box-counting fractal dimension combined with the color feature, in order to effectively identify the smoke from the forest background in color image. First, differential box dimension algorithm was used to calculate the fractal dimension value of the whole image. Second, based on the value of the image segmentation and smoke color characteristics in RGB color space, smoke region was recognized according to the split of differential box counting method regional discrimination. The algorithm reduced the number of sub-window within the coverage of the box to improve the accuracy of algorithm calculation and the speed of arithmetic operation. The changed gradation within the sub-window was proposed. The results showed that the improved differential box counting method, combined with color feature technique could accurately identify the smoke. The surface texture information was better reflected in the image and the speed of calculation is increased by nearly 50％.This method can be used for early warning of forest fires.

Key words: fractal, feature extraction, differential boxcounting, image segmentation, smoke identifieation, object recognition