In this study, a novel image-processing algorithm to identify flame regions in the foreground based on the combination of the RGB, YcbCr, and seeded region growth (SRG) algorithms is proposed. First, the conventional YCbCr algorithm is improved by incorporating the relationship between the red (R channel) and luminance (Y channel) components. Accordingly, the interfering noise corresponding to the reflective and non-reflective images can be removed. Moreover, in the case of noise-corrupted images, the interference associated with initial image segmentation can be eliminated. By estimating the centroid weight of the connected region, the seed can be automatically determined, resulting in region growth for the color-segmented images, which can facilitate fine segmentation. By analyzing the static and dynamic characteristics of a flame, the variation coefficients of the area and perimeter and the ratio of the centroid movement distance can be calculated. On this basis, a flame region can be distinguished from non-flame regions such as road lamps and candles. The experiment results indicate that the proposed method can not only be used to mitigate deficiencies of the individual algorithms that provide low accuracy, but can also be applied to simultaneously recognize the reflective and non-reflective regions to reduce interference and prevent inaccurate recognition.