Fig. 1. Flow chart of proposed algorithm

Fig. 2. Flame images with different brightness. (a)(c) Original images; (b) (d) brightness images of actual flame

Fig. 3. Original flame diagrams. (a) Non-reflective flame; (b) reflective flame

Fig. 4. Segmentation results of flame foreground region in the YCbCr color space model. (a) Non-reflective flame; (b) reflective flame

Fig. 5. Segmentation precision corresponding to different thresholds. (a) Reflective conditions; (b) non-reflective conditions

Fig. 6. Selection process of seed points. (a) Original image; (b) image segmentation; (c) centroid of connected area; (d) acquisition of seed points

Fig. 7. Improved region growing algorithm. (a) Seed point and its connected area adjacent to the pixel point; (b) merging of the initial seed point and the adjacent pixel point; (c) direction of region growth

Fig. 8. Comparison of segmentation results by the single-color models and the proposed method. (a)(e) Original images; (b)(f) RGB model; (c)(g) improved YCbCr model; (d)(h) proposed method

Fig. 9. Original images. (a) Candle; (b) light; (c) flame

Fig. 10. Segmentation results of flame and interference sources in Fig. 9. (a) Candle; (b) light; (c) flame

Fig. 11. Experimental results of area change characteristics

Fig. 12. Experimental results on the variation characteristics of perimeter

Fig. 13. Experimental results of centroid movement characteristics

Fig. 14. Experimental results of circularity variation characteristics

Fig. 15. Comparison of segmentation effects for non-reflective area by different algorithms. (a) Original image; (b) proposed algorithm; (c) threshold segmentation algorithm; (d) color segmentation algorithm

Fig. 16. Comparison of segmentation effects for reflective area by different algorithms. (a) Original image; (b) proposed algorithm; (c) threshold segmentation algorithm; (d) color segmentation algorithm