Robust Deep Learning Technique: U-Net Architecture for Pupil Segmentation
Swathi Gowroju, Aarti Aarti, Sandeep Kumar
Abstract
In many of the iris biometric applications plays a major role in tracking the gaze, detecting fatigue, and predicting the age of a person, etc. that were built for human-computer interaction and security applications such as border control applications or criminal tracking applications. In this paper, we proposed a novel CNN U-Net based model to perform the accurate segmentation of pupil. We experimented on the CASIA database and generated an accuracy of 90% in segmentation. We considered various parameters such as Accuracy, Loss, and Mean Square Error (MSE) to predict the efficiency of the model. The proposed system performed the segmentation of pupil from 512×512 images with MSE of 1.24.
Topics & Concepts
Computer scienceSegmentationArtificial intelligencePupilBiometricsIris recognitionMean squared errorComputer visionIRIS (biosensor)Image segmentationDeep learningGazePattern recognition (psychology)MathematicsStatisticsNeuroscienceBiologyBiometric Identification and SecurityFace recognition and analysisFace and Expression Recognition