KPmixer-a ConvMixer-based Network for Finger Knuckle Print Recognition

Abstract

Biometric technology is increasingly popular and has many practical applications in our lives, such as fingerprint recognition, face recognition, iris recognition, etc. In biometrics based recognition technologies, finger knuckle print recognition (FKP) has received a lot of research attention recently. This method has many advantages compared to the others such as fingerprint recognition and iris recognition. Motivated by the advantages of FKR and advances of deep learning, this paper proposes a finger knuckle print recognition model, namely KPmixer. In particular, we modify the Convmixer model using variable-size kernels to reduce the number of parameters of the model and help the model mix spatial information at various distances. At the same time, we recommend the SE+ module to increase the accuracy of FKP recognition. Moreover, we propose to use a set of effective data augmentation methods for FKP recognition. The performance of the proposed model is compared with modern CNN models such as Convmixer, Resnet18, MobileNet, and DenseNet, showing an outstanding result in terms of accuracy.