Background

This character offers an introduction to the background of the OSTR task, covering essential aspects such as open-set identification and recognition, conventional OCR methods, and their applications. First, we introduce the concept of open-set (or open-world). We discuss the different usage of open-set from various research areas and declare the OSTR task concern on either identification or recognition capability, with both seen and novel (abnormal) samples. Then, we introduce the study on open-set identification and open-set recognition to show how unknown instances are dealt with by various tasks involving recognition, including image classification, object detection, semantic segmentation, and instance segmentation. The introduction focuses on general ideas instead of implementations. Besides, the chapter also lays the background of conventional OCR methods, which can be categorized into three main frameworks: Word Level Prediction, Feature Aggregation, and Label Aggregation. Finally, we also introduce some existing studies beyond close-set text recognition before the OSTR task.