Novel Automated Method for Minirhizotron Image Analysis: Root Detection using Curvelet Transform

In this article a new method is introduced for distinguishing roots and background based on their digital curvelet transform in minirhizotron images. In the proposed method, the nonlinear mapping is applied to sub-band curvelet components followed by boundary detection using energy optimization concept. The curvelet transform has the excellent capability in detecting roots with different orientations and contrasts, thanks to its better sparse representation and more directionality feature than existing approaches. Furthermore, adapting the parameters of the mapping function due to curvelet coefficients is very beneficial for magnifying weak ridges as well as better compatibility with different minirhizotron images. Performance of the proposed method is evaluated on several minirhizotron images in two different scenarios. In the first scenario, images contain several roots, while the second scenario belongs to no-root images, which increases the chance of false detections. The results show that the detection rate of the proposed method is 4 to 27 percent better than its alternatives, in presence of zero false detection. Furthermore, it is shown that better characterization of roots by proposed algorithm does not lead to extract more false objects compared to the results of the other examined algorithms.