Pupil detection supported by Haar feature based cascade classifier for two-photon vision examinations

The aim of this paper is to present a novel method, called Adaptive Edge Detection (AED), of extraction of precise pupil edge coordinates from eye image characterized by reflections of external illuminators and laser beams. The method is used for monitoring of pupil size and position during psychophysical tests of two-photon vision performed by dedicated optical set-up. Two-photon vision is a new phenomenon of perception of short-pulsed near infrared laser beams focused on human retina as having color of half of their wavelength and it is caused by two-photon absorption occurring in human photoreceptors during safe illumination conditions. The AED method is constructed of four basic image processing operations of computational complexity of order approx. O(width*height), which makes possible to use it in real time applications. Furthermore, to achieve high resistance of AED method for presence of light reflections on images as well as other difficulties, we apply machine learning model - Haar feature based cascade (Hfbc) classifier. After implementation of trained Hfbc classifier, our method is able to find pupil edge on images, for which it was failed with this before. Finally, we obtain pupil radius and central pixel coordinates even for images containing eyelashes, changed position of illuminator or presence of reflections caused by stimulating laser beam, which significantly improves pupil edge detection efficiency.