Research Of Fire Image Segmentation And Character Selection Method

Fire is one of the major disasters that human beings are facing in current society,the occurrence of fire affects the development of social economy seriously,destroys the natural ecology,and also brings loss to human life and property.Therefore,early fire detection is of great significance.However,the traditional fire detectors using smoke,heat and radiation have the disadvantages of slow response speed,poor reliability and small detection range.At present,the research focuses in the field of fire detection is video fire detection,which is based on digital image processing technology.The technology can be used to detect early fire without space restriction,and has obvious advantages.In this paper,the existing fire detection algorithm based on image is researched,the advantages and disadvantages of these algorithm are analyzed.On the basis,the low contrast fire image segmentation algorithm is proposed,and flame image feature extraction and selection algorithm are researched.Including the following aspects:(1)The low contrast flame image segmentation.For the contrast between flame and background is not obvious,it is hard to segment the low contrast fire region.A flame enhancement and segmentation algorithm is proposed in this paper.First a color-bilateral filter is constructed by using the Retinex algorithm in the YCbCr color space,which enhances the low contrast flame image.Then,the candidate flame area is obtained by combining difference frames and color space model,and set the initial contour curve of the CV(Chan-Vese)model in the area,the further segmentation obtains the flame area.The simulation results show that the proposed algorithm can effectivelyimprove the accuracy of low contrast flame image segmentation.(2)The flame image feature extraction and selection.Based on the study of the characteristics of the fire common spatial domain,the LBP features are extracted in this paper,the DCT coefficients which by discrete cosine transform are proposed to the fire frequency domain features,and K-W(Kruskal and Wallis)checking is used to select the spatial and frequency domain features.The feature dimension selection problem is analyzed and studied,support vector machine is used to recognize the selected features.The simulation experiment shows that comparing with the algorithm that only extracts the feature of the spatial domain and select fire characteristics by artificial,the proposed algorithm has higher accuracy.