Convolutional Neural Network Image Classification Based on Different Color Spaces

This paper presents an image classification framework using Convolutional Neural Networks (CNNs) based on different color spaces, including RGB, HSV, and LAB, to enhance classification performance. The process begins with the creation of a color space images dataset, where input images are transformed into multiple color spaces. These images then undergo pre-processing steps, such as resizing and normalization, to prepare them for CNN processing. A Pseudo-Siamese CNN architecture is employed, which consists of two parallel convolutional networks, designed to learn image features from different color spaces simultaneously. This approach allows the model to leverage the unique characteristics of each color space for more effective image classification. The CNN extracts feature from the pre-processed images, and classification is performed across multiple color spaces, providing a comparative analysis of their impact on performance. The classification accuracy of the model is evaluated for various color spaces, demonstrating that certain color representations, like LAB and HSV, outperform the traditional RGB space in terms of accuracy for specific datasets. Extensive experiments indicate that utilizing multiple color spaces in image classification can significantly improve model accuracy and robustness, making this method suitable for applications in image recognition and computer vision tasks.

Index Terms—color space; Convolutional Neural Network (CNN); image classification; pseudo-Siamese network