The following two baselines will be used for this task.

Deep learning based reconstruction: This baseline will be used for reconstructing/enhancing the (degraded) fingerprint images with a standard deep neural network (DNN). To this end, a DNN is trained on the synthetic training set by minimizing a weighted combination of adversarial, feature and pixel loss functions with Adam. The DNN comprises four convolutional layers, five residual blocks, two deconvolutional layers and another convolutional layer. Each of the five residual blocks comprises two convolutional layers. All of the layers except for the last layer are followed by batch normalization and rectified linear units. The last layer is followed by batch normalization and hyperbolic tangent units. It is implemented in Chainer.

Minutiae based matching: This baseline will be used as the above mentioned publicly available standard matcher for evaluating the matching performance on the enhanced fingerprint images post submission. To this end, the NIST Biometric Image Software (NBIS) will be used. That is, MINDTCT will be used for extracting minutiae, and BOZORTH3 will be used for comparing the extracted minutiae.


ECCV Satellite Event and TPAMI

Challenge and associated ECCV Satellite Event and TPAMI special issue on Inpainting and Denoising in the Deep Learning Age!