Multi-loss Function in Robust Convolutional Autoencoder for Reconstruction Low-quality Fingerprint Image

• Main Authors: Raswa, Farchan Hakim, Halberd, Franki, Harjoko, Agus, Wahyono, Wahyono, Lee, Chung-Ting, Wang, Jia Ching
• Format: Other NonPeerReviewed
• Language: English
• Published: Proceedings of 2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA ASC 2022 2022
• Subjects: Information and Computing Sciences
• Online Access: https://repository.ugm.ac.id/284285/1/169.Multi-loss_Function_in_Robust_Convolutional_Autoencoder_for_Reconstruction_Low-quality_Fingerprint_Image.pdf; https://repository.ugm.ac.id/284285/; https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9980345
• Institution: Universitas Gadjah Mada

Our research is fingerprint reconstruction based on a convolutional autoencoder. We combine the perceptual measurement as a multi-loss function to give satisfactory weight correction, such as the structural similarity index measure (SSIM), Mean Absolute Error (MAE), and peak signal-to-noise ratio (PSNR). We observed and investigated the result using multi-loss functions and other loss functions. Eventually, our experiment obtained the highest image quality metric scores from the experimental result summarized as a loss function (SSIM + PSNR) with optimizer Root Mean Squared Propagation (RMSprop). We evaluated the image reconstruction using a dataset from FVC2004. Eventually, our proposed method gets impressive results, increasing the image's average quality by PSNR of 20.58%, SSIM of 4.07%, and MSE of 38.92%, respectively.