Validation of experiments involving image segmentation of botanic seeds

An automatic and effective procedure is proposed for validating the outcome produced by a binary image segmentation method using the CART classification algorithm and a random forests (RF) approach. It is based on criteria measuring the trade-off between classification accuracy, in particular sensitivity of a classifier, and computational complexity expressed in terms of the minimum size of the training set in experiments involving large datasets. An example from classification of botanic seeds illustrates the effectiveness of the proposed approach.