Deep Learning Models for Skin Cancer Classification Across Diverse Color Spaces: Comprehensive Analysis

Abstract

Color space plays an important role in various aspects of imaging tasks. However, in deep learning-based computer vision, the RGB color model is predominantly employed. This research analyzes the impact of deep convolutional neural networks on cancer classification across different color spaces. The five most popular deep learning models undergo training and testing in eleven color spaces, revealing that YUV, LAB, and YIQ consistently outperform other color models in most cases. RGB images are frequently converted to alternative color spaces for enhanced representation in specific applications, like object detection and segmentation. This transformation induces alterations in the features of the color image due to variations in pixel intensity information across different color models. In this research, the aforementioned principle is applied to the classification of skin cancer using deep learning networks on images of skin lesions. The results exhibit diverse responses, with some networks achieving higher accuracy in alternative color spaces compared to RGB, while others do not. This study provides insights into the classification performance across RGB, HED, HSV, LAB, RGBCIE, XYZ, YCbCr, YDbDr, YIQ, YPbPr, and YUV color spaces. The research aims to illustrate how deep learning facilitates the analysis of skin cancer images in different color spaces.