Eye pupil localization algorithm using convolutional neural networks
- PDF / 1,102,319 Bytes
- 12 Pages / 439.37 x 666.142 pts Page_size
- 103 Downloads / 225 Views
Eye pupil localization algorithm using convolutional neural networks Jun Ho Choi 1 & Kang Il Lee 1 & Byung Cheol Song 1 Received: 9 August 2019 / Revised: 16 June 2020 / Accepted: 25 August 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract
Eye pupil localization is one of the indispensable technologies in various computer vision applications such as virtual reality and augmented reality. In general, the algorithm consists of finding the approximate eye region and finding the pupil position by extracting the semantic feature from each eye region. However, the performance of the eye pupil location is affected not only by illumination and image resolution but also by glasses wear. Therefore, this paper proposes an eye pupil localization algorithm which is robust against the above disturbance conditions and provides high accuracy. First, a face is detected from an input image and it is determined whether to wear glasses using the detected face. If glasses are present, the glasses are removed to find the correct eye region. Then, facial landmarks are extracted, and eye regions are detected based on facial landmarks. Next, the pupil region is segmented using fully convolutional networks. Finally, the position of the segmented pupil is calculated. Experimental results show that the proposed algorithm outperforms the state-of-the-art algorithms for public databases such as BioID and GI4E by up to 3.44% 0.5%, respectively. Keywords Eye pupil localization . Fully convolutional networks . Eyeglasses removal
1 Introduction Eye pupil localization is a computer vision technique to locate the pupils of a person in an image. Recently, eye pupil localization has been widely used in various fields such as 3D display, intelligent robot, psychology, and brain science. This is because human eye information plays a very important role in human-computer interaction (HCI). The environment of eye pupil localization can be classified into two types. One is to detect a pupil in the eye region photographed at a close distance through a camera embedded in a head mounted device or glasses worn by the user (see Fig. 1a), as * Byung Cheol Song [email protected]
1
Department of Electronic Engineering, Inha University, Inha-ro 100, Michuhol-gu, Incheon 22212, Republic of Korea
Multimedia Tools and Applications
an example of the LPW (labeled pupils in the wild) [1, 21]. Such a photographing environment has an advantage that it is easy to detect a pupil because it provides an almost fixed eye region with high resolution. However, since the user must wear a specific device, its utilization is limited. The other is an example of the BioID [11] database in Fig. 1b. In the ordinary scenario, the pupil of a person moving freely using a non-wearable camera is detected [2, 3, 5, 6, 8, 15, 20, 22, 23]. This user-friendly environment is a general image acquisition one, which is highly likely to be utilized. However, illumination, viewing angle, resolution, and wearing glasses can be still the obstacles to accurate pupi
Data Loading...
