Accurate Pupil Detection Using the Multi Wavelet Transform (MWT) and the Hough Transform (HT)

Abstract

Historically, pupil detection plays an important role in eye tracking and gaze

estimation systems. These systems have found numerous applications in different

domains including human-computer interaction (HCI), biomedical engineering, and

clinical diagnosis of ocular diseases. An automatic eye identification system

consists of three steps: eye localization, feature extraction, and iris detection. Pupil

detection refers to the third stage of the system. Though detection of pupils seems

to be very basic and straightforward, however, different factors like varying lighting

conditions, eyelids and eyelash occlusions, and different iris and pupil color make

this process is extremely challenging. Also, the presence of specular reflection on

the cornea complicates the detection further. In the literature, many different pupil

detection techniques have been proposed that are aimed at addressing these

challenges. However, relying on one set of features to detect pupils is not adequate

because of the variations in the images. Therefore, it has been proven that by

applying multiple sets of features that are complementary to each other, a better and

more robust pupil detection performance can be achieved

In this paper, we discuss three main different pupil detection techniques using

morphology, multi-wavelet transform, and Hough transform. The main objectives

of this paper are as follows: firstly, to understand different techniques and to

investigate how the changes in the algorithms can affect the performance of pupil

detection. Secondly, to propose a comprehensive comparison between three

different pupil detection techniques. Finally, the paper concludes based on the

comparison whether there is one technique that outperforms the others. Also, it tries

to validate the proposed method by detecting and encoding the pupil data of a

human subject. This paper is organized as follows: the next section of this paper

discusses the relevant work that describes the state of the art in the area of eye and

pupil detection. Then, the methodology of all three techniques is explained in detail.

The following section discusses the experimental results and finally, the conclusion

is given. Using MATLAB 2020a, this method is applied and tested on the IIT Delhi

(IITD) iris database v1 and the Chinese Academy of Sciences (CASIA V4) iris

image database 249 persons. When compared to real-time detection speed and

steady performance, this method's center and radius detecting accuracy is high,

reaching 98% for 2268 iris on CASIA V4 picture and 99.87% for 2240 iris images

on IITD. Its speed is also acceptable.