A Novel Strategy to Enhance the Accuracy of Human Eye Pupil Detection using Hybrid Convolutional Neural Network

This study describes the implementation of Hybrid Convolutional Neural Network techniques for the segmentation of human pupils as compared with Random Forest. Recognizing and separating the pupil area in ocular images such that eye tracking can be achieved. Technique components include feature extraction, training, testing, evaluation, hybrid model selection, post-processing, and data set preparation. During preprocessing high-quality eye images are generated to identify and classifies the dataset's pupil regions. Hybrid CNN, the combination of Mednet and LSTM(Long short term Memory) model and Random Forest are compared after the image characteristics have been determined. Models are trained and adjusted using CASIA,MMU, Iris V4 dataset, and metrics like accuracy, precision, recall, and F1 score are evaluated. After the initial segmentation, the segmented pupil regions have been altered in post-processing. The accuracy of Hybrid CNN was 96.27, while that of Random Forest was 95.17. By comparing the accuracy acquired using Hybrid CNN to Random Forest, the outcome analysis indicates that Hybrid CNN performs better overall.