Image processing for applications in artificial vision

For a robot to identify objects in a particular image, it is first necessary that it can “see” them. With this aim, in artificial vision, edge detectors are normally used, i.e. computer programmes that delimit the objects in an image and define the limits between them and the background, and between the different objects themselves. Edurne Barrenechea Tartas from Pamplona has designed one of these edge detectors and, after trials with a wide-ranging set of photographs, she has shown that the results with her design are an improvement on those obtained with other detectors commonly used to date.
These results have been published in her PhD thesis, Image processing with interval-valued fuzzy sets. Edge detection. Contrast, which she recently defended at the Public University of Navarra.

Of medical and industrial interest

The operation of the designed edge detector is simple. A picture is taken, this image put through the grey scale and then introduced into the detector which, after analysing the image, produces a final picture where all the background is in black and the pixels making up the edges appear as white dots.

Once the shapes of the objects are identified, this image can be analysed by an expert system which applies rules, for example, to determine if an object is totally square, if it has a continuity line of more than x pixels or if it has a closed angle meaning- all meaning, for example, that the ensemble is correct.

These systems can have industrial applications as well as medical ones, amongst others, given that the designed shape detector works for all kinds of images. So, for example, this detector can be adapted so that the robot can detect the doors of a particular building.

In the medical field, the research team is working on a state-wide project to design and develop a reliable system for the detection of skin melanomas, thus reducing the work of the doctor by a very significant amount. With this system, using the image of a particular mole or mark on the skin, the edges are traced and compared with the progress of its shape, for example, after three months. If, in this period, the edge has been displaced by 3 millimetres, it is possible that a melanoma is involved. Monitoring this evolution by a person is very labour-intensive and costly but not for a computer system.

A new technique

The novelty of the edge detector created by Ms Edurne Barrenechea is that it is the first one based on interval-valued fuzzy sets, given that those undertaken to date have been based on traditional techniques or on fuzzy sets – but not on interval-valued fuzzy sets.

Normally, the objective of the techniques used for the edge detection of objects is the location of points where there is a variation of intensity of greys. With the technique used by Edurne Barrenechea, the edge of the objects making up an image is a set composed of pixels of the image that are associated with a sufficiently important change with respect to the intensity of grey of their neighbours. In this way, those pixels have a change of intensity associated with them with respect to their neighbours that is either null or insignificantly small, do not form part of the edge.

Thus, an edge is a set of pixels in which each pixel has an associated numerical value. This value tells us of the local variations of intensity in the area surrounding the pixel in question. In concrete, with the edge detector created by Ms Edurne Barrenechea, each element of a set has an associated interval which indicates the jump in intensity that exists between a pixel and its neighbours.

The theory of fuzzy sets has been used a lot in image processing techniques for the following reasons: fuzziness is an attribute of nature which is reflected in images; images are two-dimensional projections of a three-dimensional world and, thus, involve a loss of information; the levels of grey are considered to be imprecise constants; and, moreover, in nature, many of the definitions such as image and edge limits are vague.