Automatic Minirhizotron Root Image Analysis Using Two-dimensional Matched Filtering and Local Entropy Thresholding

An approach to automate the procedure of extracting and measuring roots in minirhizotron images is presented. By the use of two-dimensional matched filtering and local entropy thresholding, one can efficiently enhance the local contrast of the root and then extract it from the minirhizotron image. We also present several techniques for discriminating roots against extraneous objects based on their geometric features and intensity distribution properties. Once the root is detected, its length is estimated as the length of the medial axis using a more accurate length estimator based on Kimura’s method. Experimental results on a large number of images show that our automatic approach can successfully extract and measure different types of root in different kinds of soil, as well as discriminate between genuine roots and bright extraneous objects.