A new method that is 700 times faster than the norm is developed to magnify digital images

Her algorithm to magnify images stands out not only due to the quality obtained but also due to the time it takes to execute, which is 700 times less than other existing methods that obtain the same quality.

Image processing consists of a set of techniques that are applied to images to solve two problems: to improve the visual quality and to process the information contained in the image so that a computer can understand it on its own.

Nowadays, image thresholding is used to resolve many problems. Some of them include remote sensing where it is necessary to locate specific objects like rivers, forests or crops in aerial images; the analysis of medical tests to locate different structures (organs, tumours, etc.), to measure the volumes of tissue and even to carry out computer-guided surgery; or the recognition of patterns, for example to identify a vehicle registration plate at the entrance to a car park or for personal identification by means of fingerprints.
“Image thresholding separates out each of the objects that comprise the image,” explains Aránzazu Jurío. To do this, each of the pixels is analysed so that all the ones sharing the same features are considered to form part of the same object.”

The thesis entitled “Numerical measures for image processing. Magnification and Thresholding” has produced six papers, which have been published in the most highly rated journals in the field.

Super resolution

The other problem that Aránzazu Jurío has tackled in her thesis is image magnification. It involves increasing the spatial resolution of the image (to obtain a larger image with more pixels that represents the same scene) while preserving the details and sharpness. “Magnification techniques are very useful when we send images from one device to another or when we upload them to the Internet, since in order to make the transmission faster we tend to send a reduced version of the image which, when it arrives at its destination needs to be enlarged to make it available in its original size. Magnification is also used in cases in which the resolution of the image is poor, which may be the case in CCTV surveillance cameras,” she points out.

In the course of her research she has presented two new magnification methods, one for greyscale images and the other for colour images. As she points out, the methods were developed to solve a problem of an infographics company. Starting with a three-dimensional model the company used to generate various images to show to its clients; these images needed to be large so that all the details could be appreciated, but generating them took over 20 hours per image. “The solution we found meant that images could be generated in a smaller size and then enlarged in a very short space of time (less than one hour per image) while maintaining quality. In other words, our algorithm to enlarge images stands out not only due the quality obtained but also due to the time it takes to execute, which is 700 times less than other existing methods that obtain the same quality.”

Fingerprints and the human brain

In her PhD thesis this researcher has also presented two thresholding algorithms. The first is adapted to working with fingerprint images; the second is geared towards brain images obtained by means of MRI scans.

Specifically, the NUP/UPNA's research group she belongs to is collaborating on a project to create an identification centre by means of fingerprints that is capable of handling 40 million prints. “One of the steps in the identification consists of efficiently separating the fingerprint from the image background. In the thesis we proposed a way of measuring the homogeneity of each zone of the image, in other words, to see how similar all the pixels in a region are. On the basis of this measurement we have developed an algorithm that is capable of correctly carrying out the fingerprint thresholding.”

The second algorithm has been developed in the framework of a research project in collaboration with doctors at the Complejo Hospitalario de Navarra. The aim is to study the differences in the shapes or volumes of certain areas of the brain in patients who are suffering their first psychotic episodes. The researchers have come up with a method to be able to correctly separate out the area occupied by different brain structures in the image.

Reference

Humberto Bustince; Aranzazu Jurio; Ana Pradera; Radko Mesiar; Gleb Beliakov. Generalization of the weighted voting method using penalty functions constructed via faithful restricted dissimilarity functions. European Journal of Operational Research 225 – 3: 472 – 478. (2013).