The Importance of Color Spaces for Image Classification Using Artificial Neural Networks: A Review

Image classification is one of the most important applications of artificial neural networks in the field of industry and research. In most research work, when implementing ANN-based image classification models, the images used for training and testing are always represented in the RGB color space. But recent articles show that the use of certain color spaces, other than RGB, can lead to better precision in ANN-based image classification models. Thus, in this work, we present an analysis of several relevant research articles about the importance of color spaces for image classification using artificial neural networks. Thus, through the review of these articles, we will evaluate the behavior and efficiency of several ANN architectures, in different image classification contexts, and using images data sets represented in several color spaces. In the end, we not only found that there is a clear influence of the color spaces in the final accuracy of this type of tasks. But, we also found that both the creation of new special ANN architectures, and the creation of new color spaces formed from the combination of others, can lead to an increase in the performance of ANN-based image classification models.