A Multi-Task Learning Framework for TextureAware Chromatic Aberration Detection

In recent years, although methods based on traditional image processing technology have dominated the field of automated surface chromatic aberration detection of ceramic tiles, they only focus on texture patterns or color features. Faced with the increasing complexity of ceramic tile textures and colors, these traditional methods have shown significant limitations, and there are complexity issues such as the detection of chromatic aberration in tiles of the same model but with different textures. Moreover, effectively applying deep learning techniques to ceramic tile chromatic aberration detection remains a challenge. In order to solve the numerous issues encountered by traditional and deep learning methods in detecting chromatic aberration in complex textured tiles, This paper presents a multi-task learning framework for chromatic aberration detection, which includes two stages: multi-task image representation and chromatic aberration detection. In the first stage, a deep learning-based multitask model was designed to extract texture information and global color features, and its effectiveness has been proven through experiments on widely used texture datasets and real ceramic tile datasets. In the chromatic aberration detection stage, the histogram of the ceramic tile image in the CIEL* $a^{*} b^{*}$ color space and the mean and variance of the corresponding template ceramic tile image (flag features) are extracted and fused with the features extracted in the first stage, forming a descriptor that represents the ceramic tile image. This image descriptor will be used in the training of the Multilayer Prceptron (MLP) to obtain a classifier for chromatic aberration detection. Experiments on real ceramic tile datasets have shown that this chromatic aberration detection framework can efficiently and accurately detect defective products with chromatic aberration.