Flame image segmentation using multiscale color and wavelet-based texture features

Accurate and reliable segmentation of flame images are crucial in vision based monitoring and characterization of flames. It is, however, difficult to maintain the segmentation accuracy while achieving fast processing time due to the impact of the background noise in the images and the variation of operation conditions. To improve the quality of the image segmentation, a flame image segmentation method is proposed based on Multiscale Color and Wavelet-based Textures?MCWT? of the images. By combining the color and texture features, a characteristic matrix is built and then compressed using a local mean method. The outer contour of the flame image under the compressed scale is detected by a cluster technique. Subsequently, the flame edge region under the original scale is determined, following that, the characteristic matrix of the edge region is constructed and classified, and finally, the flame image segmentation is achieved. Flame images captured from an industrial-scale coal-firedtest rig under different operation conditions are segmented to evaluate the proposed method. The test results demonstrate that the performance of segmenting flame images of the proposed method is superior to other traditional methods. It also has been found that the proposed method performs more effectively in segmenting the flame images with Gaussian and pepper and salt noise.