Industrial engineer Laura Fernández Robles has developed software solutions for three different industrial partners requiring image recognition: pig farmers, manufacturers and the Spanish police. By combining different approaches, she was able to help the pig farmers automate artificial insemination, the manufacturers inspect milling tools and the Spanish police fight child pornography.
In a joint PhD programme at the University of León (Spain) and the University of Groningen, she created a tool that has been implemented by the Spanish police. The PhD ceremony in Groningen will be on 5 Februar 2016.
It is a daunting task for the police to establish links between cases when faced with thousands of pictures of abused children. It would help if they could see which images came from the same source, on the basis of objects like toys. ‘But there was no automated system to analyze the images’, says Fernández Robles.
The software she created extracts the features of objects in an image and stores these in a database. ‘The idea is that police officers select an object in one image, and the system then checks the database for similar objects’, says Fernández Robles. The software she used is able to match partial descriptions, so it can deliver a result even if only part of an object is visible. ‘It can also recognize an object even if the viewpoint differs between images’, she says.
Fernández Robles used different description methods and filters to classify the objects. ‘One of my contributions is the addition of colour information to one of the filters, the COSFIRE filter.’ She did some of the work in the Intelligent Systems research group of her supervisor Nicolai Petkov at the University of Groningen and the rest of the work in Spain.
A particular challenge was the large number of pictures the police have to contend with. To speed up comparison, all these pictures need to be analyzed by the software and the pre-computed descriptors stored in a database. The Spanish police are now in the process of uploading images to the system. Fernández Robles says, ‘They were happy with the first results and will be using the system on a larger scale in the near future.’
The work on child pornography was part of the ASASEC European research project, which is coordinated by INCIBE, the Spanish National Cybersecurity Institute. In her thesis, Fernández Robles describes two more projects involving industrial partners. In the first project she developed an automated system to assess the quality of boar sperm, and in the second a system to inspect the integrity of inserts in industrial cutting tools. ‘In contrast to the police pictures, these two projects involved known objects, so I built systems that could be trained.’
Pig farmers are already using the software to check the quality of sperm to be used for breeding. ‘It is now an automated system commercialized by the company Microptic SL, whereas previously the quality of the spermatozoa had to be assessed in a visual inspection through an expensive fluorescence microscope.’ The system to monitor cutting tools for degradation is working well in the lab, but needs to be tested in a real factory. Fernández Robles will continue her research and do some lecturing at the University of León: ‘Being an engineer, I do like real-world applications.’
This research has been supported by the University o f León through a teaching assistant contract, the Junta de Castilla y León through a grant destined to fund the hiring of recently graduated research personnel, the Spanish government through project grants DPI2009-08424 and PR2009-0280 for boar sperm assessment and DPI2012-36166 for inserts inspection and the European Union through grant HOME/2010/ISEC/AG/043 for the ASASEC project.
Laura Fernández Robles will defend her PhD thesis entitled
‘Object recognition techniques in real applications’
on Friday 5 February at the University of Groningen. Her promotor in Groningen is prof.dr. Nicolai Petkov.
On January 15th
Herib conducted his research at the Center for Energy and Environmental Sciences - IVEM
In memory of Prof. Dr. Albert van Leusen