<p>Contactless palmprint recognition technology enable reliable personal recognition by using less-constrained and applicable procedures where the palm are not required to touch with a surface. To achieve accurate recognition, various techniques like local texture descriptors and convolutional neural networks (CNNs) are recently proposed to explore representative features while making up for variations in rotations, translations, scales, and illuminations. However, the available CNN-based techniques focus on either sophisticated designs, which need numerous manually-tuned parameters, or pretrained filters produced by general-purpose datasets, which could not be particularly appropriate for palmprint processing. In this paper, we propose GaborNet, which is a novel method that applies Gabor filter in a CNN and rescales Gabor responses through attention mechanism, and uses multiple direction binary coding strategy to construct high discriminative descriptors for contactless palmprint recognition. Experimental results on several palmprints datasets captured under different acquisition procedures and distinct devices demonstrate that, our method could obtain recognition accuracies greater than that of the state-of-the-art techniques.</p>