Convolutional Autoencoders for Image Comparison in Printing Industry Quality Control

Abstract

An important problem in the printing industry is the automation of the quality control of the printed materials and timely detection of relevant printing faults. Image quality assessment can be used to support the quality evaluation of the printed material based on comparison of initial input image and a camera captured image of the printed result. Besides the utilization of histogram based analyses and statistical evaluation metrics such as mean squared error and structural similarity index, deep learning techniques can also be applied for image comparison. The paper presents a deep learning approach based on convolutional autoencoder for image comparison utilizing data augmentation and clustering in order to evaluate the quality of printed materials.