Name
Abstract | ||
|---|---|---|
With the increasing demand for identity authentication, iris recognition systems are gradually applied in various scenarios. Iris segmentation is the basis of the iris recognition system. Iris images taken in non-ideal situations contain much irrelevant noise. In addition, the quality of iris images varies greatly under different shooting conditions. These factors seriously affect the accuracy of iris segmentation. However, traditional algorithms are not adaptable enough, and the algorithms based on convolutional neural networks (CNNs) are not efficient enough. Inspired by the success of U-shaped network++ (UNet++) in image segmentation, in this paper, an end-to-end encoder-decoder model based on improved UNet++ is proposed to perform the iris segmentation, referred to as Attention Mechanism UNet++ (AM-UNet++). The main contributions are as follows: firstly, EfficientNetV2 is selected as convolutional blocks of UNet++ to improve training speed and reduce the number of network parameters. Secondly, an attention module is embedded into the down-sampling process of UNet++ to suppress irrelevant noise interference and strengthen the ability to learn the discriminability of the iris region. Finally, the algorithm adopts a pruning scheme to get four different performance networks, which can meet the needs of iris recognition in multiple scenarios. The experimental results on two near-infrared illumination iris databases and one visible light illumination iris database demonstrate that the method has good iris segmentation ability and generalization performance. The iris segmentation accuracy and efficiency of the proposed method are higher than the state-of-the-art fusion method. |
Year | DOI | Venue |
|---|---|---|
2022 | 10.1007/s11042-022-13198-z | MULTIMEDIA TOOLS AND APPLICATIONS |
Keywords | DocType | Volume |
Iris segmentation, Iris recognition, CNNs, UNet plus plus | Journal | 81 |
Issue | ISSN | Citations |
28 | 1380-7501 | 0 |
PageRank | References | Authors |
0.34 | 0 | 3 |