A real-time method to extract weed or crop populations was presented, based on RGB color components operation combined with bit-masked color reduction. Experimental results on different weed or crop populations show that color components operation can enhance the visual pattern of objects to human, and color bit-masked operation can reduce noises in image, so that the error on segmentation can be reduced. Processing time of this method costs little, no matter how complex of related scenario, with color objects be extracted within 30 ms for images of 320×240 resolution, and it's robust to variation of color or illumination on local weed or crop populations. The coefficient and operator for color components combination is determined by a supervised learning mechanism.