Identification Systems

Due to its excellent user-friendliness, including its non-invasive and hygienic features, palmprint recognition has attracted a lot of study interest. Most current palmprint identification works, such as Deep Learning (DL) techniques, usually learn discriminative attributes from palmprint pictures. These typically require significant labeled samples for an acceptable recognition performance. However, the availability of palmprint images is typically constrained due to the challenge of gathering sufficient samples, making most deep learning-based algorithms unsuccessful. To overcome the issues, we propose a deep-learning approach to palmprint recognition. Initially, (a) convert the original image to greyscale, (b) crop and resize the image, and (c) enhance the contrast utilizing Contrast-Limited Adaptive Histogram Equalization (CLAHE) approach. After that, extract the salient features from contrast-enhanced palm images using the ConvNeXt technique. Then, using Improved Spotted Hyena Optimizer (ISHO) algorithm to select the essential features by eliminating unnecessary extracted features. Finally, we employ the Darknet Convolutional Neural Network (DNetCNN) approach to classify whether the palmprint image is matched or not-matched. To improve the classification accuracy, utilize Sooty Tern Optimization Algorithm (STOA) to get higher classification accuracy and recognition accuracy. The proposed approach efficiently increases the Optimized DNetCNN's accuracy while decreasing the number of features and computing time. We validated the proposed methodology on three open palmprint databases, CASIA, Tongji, and IITD. A high recognition rate is achieved by our approach while utilizing a significantly smaller number of characteristics, according to experiments.