Segmentation of rice canopy image using the Otsu method based on visual spectral image color related indices

Digital image analysis of rice canopy has been widely for monitoring rice growth，diagnosis rice N status，and predicting rice yield. The accuracy，stability，and reliability of the rice canopy digital image analysis result were based on the premise of precise segmentation of rice pixels. Otsu’s method was the frequently used thresholding method in image segmentation，for its stability and self-adaptation. This paper extracted and calculated 9 image color indices from digital camera images using three color models： R，G，B，L*，a*，b* components of CIEL*a*b* color space，H，visible-spectrum green leaf algorithm（ GLD） and vegetation index（VIGreen）. The segmentation of rice pixel from soil background was conducted by the Otsu’s method based on different color indices，and the accuracy of the segmentation was evaluated. The results showed that there were obvious bimodality and little overlap in a*，b* components of CIEL*a*b* color space，GLD，and VIGreen between rice and soil pixels， which mean that these color indices should be the candidate color indices for rice canopy image segmentation. The Otsu’s method based on a* components of CIEL*a*b* color space，GLD，and VIGreen had higher segmentation accuracy in comparison with which based on b* components of CIEL*a*b* color space. The highest signal-noise ratio and lowest error rate of rice canopy image segmentation was arrived by the Otsu’s method based on a* components of CIEL*a*b* color space. The signal-noise ratio and error rate of rice canopy image segmentation by the Otsu’s method based on VIGreen were next only to the Otsu’s method based on a* components of CIEL*a*b* color space. In conclusion，a* components of CIEL*a*b* color space was the optimum image color index for rice canopy image segmentation by the Otsu’s method.