Object Detection과 OCR을 활용한 다중 책 제목 검출 모델

다국어 초록 (Multilingual Abstract)

In this paper, we discuss the results of implementing a model that uses a combination of object detection and Optical Character Recognition(OCR) to detect book spine and extract book titles when multiple books in a bookcase are plugged in, thereby confirming additional information about the book. The object detection process was divided into book spine detection and book title area detection, and two Yolov5 models were used to increase the performance of object detection. As an OCR result, pre-processing such as grayscale, resizing, and brightness adjustment was performed before entering images into OCR model to increase the accuracy of text output. In the process of applying a model designed to compensate for the problem of slow OCR processing time to web applications, it has added a function that allows users to select the book they want. In addition, information such as author, publisher, price, and reviews can be confirmed through book titles extracted through object detection and OCR.