A new color augmentation method for deep learning segmentation of histological images

Abstract : This paper addresses the problem of labeled data insufficiency in neural network training for semantic segmentation of color-stained histological images acquired via Whole Slide Imaging. It proposes an efficient image augmentation method to alleviate the demand for a large amount of labeled data and improve the network's generalization capacity. Typical image augmentation in bioimaging involves geometric transformation. Here, we propose a new image augmentation technique by combining the structure of one image with the color appearance of another image to construct augmented images on-the-fly for each training iteration. We show that it improves performance in the segmentation of histological images of human skin, and also offers better results when combined with geometric transformation .