• Home
• News
• Organized events
• Location

• Staff
• Students
• Technicians
• Visitors
• Former members

• Databases
• PhD Theses
• Projects
• Publications
• Software

• Access

Computer Vision

Deep learning techniques have shown a great ability in classification tasks when dealing with images, and generally with signals. Particularly, Convolutional Neural Networks have recently achieved a huge performance leap for computer vision tasks.

The main worklines of our group are:

• Image classification. One of the main computer vision projects in our group is MirBot (http://www.mirbot.com), a Multimodal Interactive Image Information Retrieval system based on convolutional neural networks. The MirBot app can be downloaded for smartphones, and allows users to classify images. This app is used to collect a large image dataset of images, with their regions of interest, minimal occlusions, and labelled with WordNet synsets. Besides, we are also working on neural networks applied to other problems such as ship detection from aircraft images, and OMR (Optical Music Recognition),

• Image similarity. Convolutional neural networks achieve an excellent performance in classification, but their accuracy for image retrieval (image similarity or instance-based) tasks still has room for improvement. We are also working on neural network topologies applied to image retrieval problems.

Related Links:

MirBot (http://www.mirbot.com)