Self-Supervised Learning Framework toward State-of-the-Art Iris Image Segmentation

Wenny Ramadha Putri,Jia-Ching Wang,Muhammad Saqlain Aslam,Chin-Chen Chang,Yung-Hui Li,Shen-Hsuan Liu

doi:10.3390/s22062133

Abstract

Iris segmentation plays a pivotal role in the iris recognition system. The deep learning technique developed in recent years has gradually been applied to iris recognition techniques. As we all know, applying deep learning techniques requires a large number of data sets with high-quality manual labels. The larger the amount of data, the better the algorithm performs. In this paper, we propose a self-supervised framework utilizing the pix2pix conditional adversarial network for generating unlimited diversified iris images. Then, the generated iris images are used to train the iris segmentation network to achieve state-of-the-art performance. We also propose an algorithm to generate iris masks based on 11 tunable parameters, which can be generated randomly. Such a framework can generate an unlimited amount of photo-realistic training data for down-stream tasks. Experimental results demonstrate that the proposed framework achieved promising results in all commonly used metrics. The proposed framework can be easily generalized to any object segmentation task with a simple fine-tuning of the mask generation algorithm.

Highlights

Over the past few years, iris recognition has emerged as one of the most suitable and trustworthy biometric modalities among those currently available in the private sector [1–4]
We introduce an improved version of the Pix2Pix-based conditional adversarial generative model, which can serve to generate a vast amount of iris images with pre-defined iris masks and periocular masks
Since there is no objective evaluation method to examine whether the generated image is true, and our goal is to improve the deep learning algorithm for iris segmentation task by hallucinating training data, we assess the performance of generated images by analyzing the segmentation accuracy of the down-stream segmentation networks

Summary

Introduction

Over the past few years, iris recognition has emerged as one of the most suitable and trustworthy biometric modalities among those currently available in the private sector [1–4]. The iris recognition system has been operating globally, and it represents one of the most developed categories of biometric recognition technology [10]. It can solve technical obstacles when face recognition is failed or unavailable, peculiarly when the user’s face is covered by masks, especially in the COVID-19 era. The accurate iris segmentation, combined with the best features and effective recognition schemes, makes the iris recognition system more perfect. If the iris segmentation is not accurate, the best feature extraction and recognition algorithms cannot compensate for such defects. With the rapid development of deep learning, a vast number of investigations employing CNNs have been introduced for iris segmentation [1,16–20], iris bounding box identification [19], and pupil center identification [21–23]

Methods

Results

Conclusion