Whiteboard content extraction and enhancement for videoconferencing systems

Abstract

Whiteboards have become essential for teaching, presentations and conferences since they are very flexible tools for spontaneous knowledge sharing. It’s content can be captured by videoconferencing systems in order to create more engaging presentations to remote users. However depending on the lighting configuration in the scene and the position of the camera with respect to the whiteboard some shading effects and highlights may appear decreasing the contrast and legibility of the captured images. We propose a system that records and extracts the whiteboard content and enhances its appearance and legibility. Our system acquires a sequence of images from a high resolution fixed view camera and extracts the whiteboard content. Firstly we estimate the whiteboard background model using a robust surface fitting technique. We use this model to locally identify areas with written content and also to detect and remove occlusions. Secondly we estimate the illumination axis that goes through the whiteboard color distribution in the RGB color space. Finally we implement two different enhancement methods: one that balances the image colors rotating the illumination axis and another focused on enhancing the image colors using the estimated color of the illuminant. We perform a psycho-visual experiment on a dataset of images enhanced with our methods combined with other state of the art whiteboard image enhancement algorithms. The experimental results shows that our fist enhancement method provide statically more visually appealing images while our second method provide statically more legible images.