Researchers at Max Planck Institute for Informatics (MPII) have developed a video inpainting software that can effectively remove people or objects from high definition footage. Inpainting is the process of reconstructing lost or deteriorated parts of images and videos. When people or objects are digitally removed from an image, it will leave an empty space in the background. Inpainting fills up the empty space with parts of image or video frame with the same background.
The software analyzes each video frames and determines the pixels that need to be removed. It also determines the pixels that should replace the area marked for removal. In other words, the software can compensate for multiple people overlapped by unwanted element, even if they are walking towards or away from the camera. From the camera’s point of view, when a person moves towards the camera, his position does not change. He only appears bigger as the scene progresses. Likewise when the person moves away form the camera, he appears smaller as the scene progresses.
The software was developed by Graphics, Vision and Video research group at MPII, led by Prof. Dr. Christian Theobalt and Jan Kautz, Professor of Visual Computing at University College London. It is a big improvement in inpainting method which previously only worked with low resolution video but produce poor results with high definition videos. The new software still generates some artifacts, but they are almost imperceptible to untrained eyes.
The software works best with video with static background. Similar technique had been employed to remove people from Google’s Street View for privacy concerns.