Comparison of threshold algorithms for automatic image processing of rice roots using freeware ImageJ

High-throughput phenotyping of crop root system architecture using digital image analysis requires costly special software (WinRHIZO). Novel freeware, ImageJ, has also been developed for the similar purpose, but the presetting of the threshold to create root binary images makes this method time-consuming. This study compares the 16 algorithms available in the ImageJ for processing of rice root images. Among the algorithms, the Triangle algorithm proved to be the best binary method, where the coefficient of correlation between ImageJ-estimated and WinRHIZO-estimated root lengths is extremely high (r = 0.986). However, using the Triangle algorithm, ImageJ overestimated the rice root length compared with WinRHIZO. By multiplying the values obtained using ImageJ by 2/3, the estimates closely corresponded to those estimated by WinRHIZO. The correspondence of root lengths estimated using WinRHIZO and ImageJ with Triangle algorithm was valid for roots of various morphologies, and for rice plants grown in uplands, rainfed lowlands, and irrigated lowlands. This report proposes the completely automated estimation of rice root length using freeware ImageJ with the appropriate threshold algorithm for image processing.