Fingerprints are the marks left by the friction of human fingers. Fingerprint identification compares two friction skin impressions of human fingers or toes to resolve whether the impressions may have come from the same person. Fingerprint identification is one of the biometric identification technologies. Most mainstream fingerprint identification algorithms are based on artificially predefined traditional fingerprint features, such as orientation field and singular point. However, because these algorithms rely heavily on artificial predefined features, when the image noise is large, the artificial predefined features cannot be extracted effectively, resulting in poor identification accuracy. After 2012, some researchers used convolutional neural networks for fingerprint identification tasks and achieved higher accuracy. However, this paper believes that the fingerprint images have relatively special curves, so the traditional convolutional neural network algorithms cannot fully extract the characteristics of the curves. Therefore, this paper proposes the siamese rectangular convolutional neural networks (SRCNN) algorithm to solve this problem. This paper presents the detailed architecture of SRCNN and experiments on NIST Special Database 4. The experimental results show that the accuracy of SRCNN is 4% higher than that of traditional convolutional neural networks on the test set. That is to say, SRCNN performs better than traditional convolutional neural networks in fingerprint recognition tasks. Keywords—Fingerprint identification; Siamese; Rectangular Convolutional Neural Network

No abstract

No abstract

No abstract

This paper proposed an enhanced data handling technique by introducing multiple layers of security in securing sensitive text data, which is realized by using both cryptography and steganography techniques. The proposed method is cost-effective as the data compression technique is observed. This study uses Polybius cipher to transform plaintext containing vital information into an unintelligible format called ciphertext. The ciphertext is compressed using the Huffman coding algorithm, where output is embedded in an image file using the Least Significant Bit (LSB) algorithm for the steganography technique. Simulation results show that the proposed methodology produced stego images with better performance as to file size, Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), and error metrics as against the stego image generated using the lone LSB steganography technique.