Abstract—In order to improve recognition accuracy and efficiency for real-time monitoring, the paper proposed a static-dynamic fusion feature method by pixel’s color moment and covariance matrix descriptor in the CYMK and YCrCb color space. It employs the covariance matrix descriptors to represent the singular flame’s feature in a assemble vector by analyzing rationality of selection and combination of the features. It proposed a method of blending and classification of flame’s attributes by sparse dictionary from the covariance matrix descriptor and color’s low order moments. It had implemented and analyzed sparse representation classification of flame regions by MP and OMP processing. The experiments showed that the OMP sparse representation classification gain higher classification accuracy than other traditional classification methods.

Index Terms—Covariance matrix descriptor, image processing, orthogonal matching pursuit, sparse dictionary.

The authors are with East China Jiaotong University, Nanchang 330013, Jiangxi, China.

Cite:Xiangang Jiang, Chuanxiu Hu, Zizhu Fan, Panpan Zhang, "Research on Flame Detection Method by Fusion Feature and Sparse Representation Classification," International Journal of Computer and Communication Engineering vol. 5, no. 4, pp. 238-245, 2016.