A comprehensive survey of image synthesis approaches for Deep Learning-based surface defect detection in manufacturing

Detection of manufacturing defects is a crucial step in ensuring product quality and safety. The automation of defect detection processes and the enhancement of detection accuracy are pivotal objectives in industrial quality control. However, the complexities of manufacturing processes present significant hurdles in the development of effective defect detection models. Deep Learning (DL) models have emerged as a potential solution for defect detection by learning patterns from extensive datasets without necessitating an in-depth understanding of the manufacturing processes. However, training such DL models requires vast amounts of data, which are often difficult and costly to collect from real manufacturing environments. As a response to these challenges, researchers have proposed synthetic image generation to facilitate DL model training. The existing literature primarily focuses on two main approaches for synthetic defect image generation: computer graphics-based methods and DL-based methods. However, there are a limited number of literature reviews focused on DL-based methods and no reviews on recent developments particularly diffusion models in defect image synthesis. Moreover, no comprehensive review currently addresses the application of computer graphics-based techniques for defect image generation. Therefore, this article presents a comprehensive review covering both computer graphics-based methods and recent developments in DL-based methods employed in the synthesis of artificial images. The review addresses various techniques, their strengths and limitations, and their implications for advancing defect detection in manufacturing.