IMAGE PROCESSING BASED SHIRT SIZE TESTING WITH HU MOMENTS

Abstract

This study proposes a hybrid classification model for the automatic detection of shirt sizes in textile quality control processes. The model integrates Hu moment-based numerical features with a convolutional neural network (CNN) architecture, combining the interpretability of traditional image processing with the classification performance of deep learning methods. The proposed CNN+Hu moments approach was trained and tested on a controlled dataset comprising various shirt size categories (XXS, XS, S, M, L, XL). Experimental results demonstrate that the hybrid model significantly outperforms classical machine learning algorithms such as Support Vector Machines (SVM) and KDTree in terms of accuracy and robustness. With its low cost, explainability, and effectiveness, this approach offers a new dimension to quality control systems in the textile industry.