MS-Net: Multi-Segmentation Network for the Iris Region Using Deep Learning in an Unconstrained Environment

Iris segmentation is a significant phase in the iris recognition process because errors cascade into all subsequent phases. Therefore, it important that are minimised. The U-Net architecture uses deep learning approach was previously adopted for this task, but its performance affected by deformation of images caused various noise factors unconstrained (non-ideal) environments. Scratches, blurriness, dirt, specular reflections and other some challenges faced environments when eyeglasses present original images. Additionally, degraded due to problems exploding gradient or vanishing loss information. This paper proposes multi-segmentation network called MS-Net, based on approach, aims capture high-level semantic features while maintaining spatial information improve accuracy segmentation. MS-Net consists three principal segments: feature encoder network, multi-scale context extractor (MSCFE-Net) decoder network. MSCFE-Net constructed from dilated residual multi-convolutional module pyramid pooling model an attention convolutional module. In addition, proposed contains dense connections within decrease training difficulty, using only few samples. evaluated CASIA-Iris.V4-1000 UBIRIS.V2 databases. our method databases achieved overall 97.11% 96.128%, respectively. Experiment results show able achieve better compared earlier methods used same purpose.