Biometric object or facial detection, identification and recognition system using python and deep learning models
Ghanshyam Yadav, Jitendra Mohan Giri, Ravi Kalra, Irfan Ullah Khan, Pradeep K. Chandra, Abhishek Kaushik, Ashish Parmar · 2024
This research paper is focused on creating an automated attendance system for colleges, institutions, organizations and universities with the help of image processing, deep learning, objects or facial detection, identification and recognition techniques. Many colleges, institutions and organizations have implemented object or facial detection, identification, recognition and attendance system but often it fails in different lighting conditions and very often give false positives. Many false positives also happen because objects or facial detection, identification and recognition techniques cannot differentiate between identical twins or in case of change in appearance. To tackle these issues, HOG (histogram of oriented gradients) algorithm as feature identifier (for the face detection phase), descriptor with SVM (support vector machine) algorithm and recognition phase by CNN (convolutional neural networks) deep learning model. HOG works much better with SVM and gives higher accuracy in different lighting conditions. This research paper still cannot differentiate identical twins, but this research paper has found a way to take their attendance using multiple cameras and at different angles and also in case of change in appearance, hairstyle [ 5 ][ 7 ]. This paper delves into the fascinating world of image identification, classification, and recognition. We’ll be harnessing the power of advanced CNN with the ever-reliable TensorFlow Framework. Python takes center stage here, thanks to its impressive array of libraries such as Dlib, openCV, and TensorFlow. CNN proves to be the ultimate technique for training and testing data, consistently delivering outstanding results in facial detection, identification, and recognition. Accuracy, precision, recall, and time-space complexity are all carefully evaluated. With advanced CNN models, we’re achieving remarkable results of over 98%, leaving others in the dust with their subpar sub-90% accuracy.