We conduct research into computer-based image- and vision-processing technologies for images, videos, and 3D and multi-view content. Together, the technologies we have developed enable:
- The automation and speeding-up of production and post-production processes: the algorithms we have researched help to automate the production and post-production processes for audiovisual content, thus saving time and production costs.
- The development of new, creative possibilities in production and post-production, thanks to the technologies for processing 3D and multi-view images and 3D reconstruction.
- Easy integration with other equipment and software tools via the integration of algorithms and plug-ins.
- The development of new applications for the media and other industries that can benefit from the use of technologies for the detection and analysis of images and video.
Detection of objects and characteristics
Development of methods to detect and identify objects in images and videos that are resistant to changes in lighting, along with transformation and degradation of the object image in relation to the sample. In particular, algorithms have been developed to detect logos in videos that also make it possible to obtain data regarding their properties (size or position within the image).
Segmentation and following of objects in videos
Research into interactive (supervised) methods of segmentation with minimum user interaction, and methods of following an object based on using the movement-related information drawn from the video in order to locate the successive positions of the object throughout a sequence.
Image, video and 3D inpainting
Development of inpainting algorithms that enable the automatic filling-in of the damaged or unwanted parts of an image, using appropriate texture so that the reconstructed image appears natural; development of inpainting extensions for application in video sequences and 3D and multi-view content which take on the additional challenge of generating coherent textures between consecutive frames and different viewpoints (particularly in the pair of images that form a 3D image).
Multiview-image processing: depth calculation, 3D scene reconstruction and synthesis of new viewpoints
Research into obtaining maps of depth and 3D information for a scene based on multiple images captured from different viewpoints. This information can be used for 3D reconstruction of the scene and for generating virtual views of the scene from new viewpoints, for both still and moving sequences, via depth-based image-rendering techniques. We also conduct research into the processing of aerial images for the automatic generation of genuine orthophotos and the reconstruction of roof surface areas.
3D editing: inserting virtual objects into 3D video
Research into technologies used to insert virtual objects into 3D video sequences in such a way that the objects are perfectly integrated into the scene and are lit correctly (precisely controlling the shadows generated), in addition to ensuring that the position, orientation and scale of the objects are consistent with the scene.
Estimating scene lighting
Research into algorithms for estimating the lighting of a scene based on images captured previously using image-analysis techniques.
Multigrid and narrow-band techniques for image processing
The objective of multi-resolution, multigrid and narrow-band techniques is to speed up certain image-processing algorithms. These techniques are applied in segmentation, 3D reconstruction and the calculation of disparity and depth.
Implementing parallel algorithms in GPU
Designing parallel algorithms in GPU for images in general and in mutigrid- and narrow band-based algorithms in particular. At present, whenever the algorithm can be made parallel the team is implementing it in GPU using the CUDA language.