부산대학교 도서관

Recently, COVID-19 has accelerated the non-contact culture. Many presentations, such as workshops and conferences, are conducted in an online and offline hybrid mode in a conference room. In presentations, a screen of the slide is particularly important. Therefore, we propose an algorithm that detects the screen in an image. Firstly, a screen region is extracted using a deep learning-based instance segmentation method. However, this extracted region has a noisy boundary. We designed an image processing algorithm composed of 7 main steps to solve this noise and detect the screen. To validate the proposed method, a real dataset was qualitatively evaluated, and the result images show that only meaningful screen regions in the test image can be extracted.