The Influence of Image Morphology on Neural Network-Based Segmentation Results

Abstract

In this article, we have provided initial results of evaluating the influence of training data morphology on convolutional neural networks segmentation quality. To do this, we selected 4 different soil samples and segmented them using an unsupervised converging active contours algorithm. With the help of synthetic tomography algorithm, we obtained a true CT– ground-truth pairs for these soil samples. Then we acquired a set of samples with varying degrees of morphological properties similarity with the original soil samples by stochastic reconstructions. In order to check the effect of morphological properties on the quality of segmentation, we trained an U-Net model with ResNet50 encoder on pairs of synthetic CT – ground-truth data from the initial soil samples, and assessed the segmentation quality of synthetic CT obtained from stochastically reconstructed samples. Based on the metrics, we concluded that the quality of segmentation is more influenced by the morphological differences of the original soil samples than by difference from the generated stochastic reconstructions. We discussed possible ways to improve the future experiments design in order to finally close the issue outlined in this work.