The Impact of Using Different Color Spaces in Histological Image Classification using Convolutional Neural Networks

Ronny Velastegui, Marius Pedersen

Abstract: Classification is an important aspect of medical image analysis. Nowadays, Convolutional Neural Networks (CNNs) are extensively used in the field of medical image classification. There are several kinds of research on medical image classification combining different CNN architectures and data sets. When using color image data sets, most of those research works use RGB as the standard color space to train and test the models. While RGB is a standard color space to represent images on multimedia devices, RGB might not be the best color space to train CNN models for medical image classification applications. We implement an AlexNet CNN to classify colon tissue images to detect tumors. We perform this task using several color spaces, such as RGB, XYZ, CIELAB, HSV, and YCbCr. We analyze the results and indicate which color spaces give the best accuracy in performing this medical image classification task.

External IDs:dblp:conf/euvip/VelasteguiP21