On the Use of an Hyperspectral Imaging Vision Based Measurement System and Machine Learning for Iris Pigmentation Grading

Nowadays, the ability to derive accurate measurements from images, i.e. the application of vision-based systems to the measurement field, is becoming an attractive research field. In this context, Machine Learning (ML) algorithms can be exploited to smartly and automatically perform the measurement activity. This paper presents an interesting application of ML techniques to an Hyperspectral Imaging System, devoted to the analysis of the iris pigmentation. Indeed, it is proven that the iris pattern evaluation gives a chance for the analysis of both possible loss of sight and future outbreak of several eye diseases. The proposed Vision-Based Measurement system (VBM) allows to illuminate the subject eyes in the spectral range 480 - 900 nm. In particular, the imaging system foresees to take 22 different images of 2048 x 1536 pixels, thus obtaining a spectral resolution of 20 nm and a spatial resolution of 10.7 mu m. In this paper, as a first research step, we evaluate the possibility to develop a suitable Machine Learning algorithm to classify the iris color. In particular, the goal is to point out the possible ML techniques that can be employed, the needed dataset and the possible advantages offered by the hyperspectral approach, compared to the conventional visible light imaging.