Classification Algorithm of 3D Pattern Film Using the Optimal Widths of a Histogram

Abstract

A recent development in marketing tools aimed at boosting sales involves the use of three-dimensional pattern films. These films are designed to captivate customers’ interest by applying 3D patterns to the surface of products. However, ensuring the quality of these produced 3D films can be quite challenging due to factors such as low contrast and unclear layout. Furthermore, there has been a shortage of research on methods for evaluating the quality of 3D pattern films, and existing approaches have often failed to yield satisfactory results. To address this pressing issue, we propose an algorithm for classifying 3D pattern films into either ‘good’ or ‘bad’ categories. Unlike conventional segmentation algorithm or edge detection methods, our proposed algorithm leverages the width information at specific heights of the image histograms. The experimental results demonstrate a significant disparity in histogram shapes between good and bad patterns. Specifically, by comparing the widths of all images at the quintile of the histogram height, we show that it is possible to achieve a 100% accuracy in classifying patterns as either ‘good’ or ‘bad’. In comparative experiments, our proposed algorithm consistently outperformed other methods, achieving the highest classification accuracy.