A DEEP CONVOLUTIONAL FRAMEWORK FOR REAL-TIME FACE RECOGNITION USING FACENET AND TRIPLET LOSS

Abstract

Face detection and recognition have become vital innovations in identifying human faces under varying condition, diverse domain application such as security system. Numerous methodologies proposed for face recognition, with achieving high accuracy remain core challenge. The proposed modality of this paper introduce the frame work on basis of FaceNet, a robust and efficient system designed to map face region image into compact Euclidean space, where distance reflect facial region similarities. Utilizing deep convolution neural network, Face Net directly optimize embedding without relying on intermediate bottleneck layers, surpassing the limitation of earlier approaches. The process of training incorporates an advancement in online triplet mining strategy, leveraging triplets of matching and non-matching face patches to optimize the triplet loss function. This technique enables the proposed modality with face net frame work to achieve high efficient embeddings of only 128 bytes per face while maintaining state-of-the-art performance. Experimental results demonstrate achieve an accuracy of 89.66% on the widely-used labeled Face in the Wild (LFW) dataset and 98.63% on an internal data set. However, the internal dataset’s characteristics and its comparability to established benchmarks are not detailed, which the limit the generalizability of these results. The frame work Face Net showcase the significant advancement in face recognition, including high accuracy, compact features representations, and adaptability to real world conditions. Future work will explore its performance across diverse datasets and investigate potential limitations and ensure broad applicability and ethical deployment.