Deep Learning Based Facial Obfuscation Using MobileNet

EasyChair Preprint 10033

Abstract

More and more daily tasks are being completed electronically rather than with pen and paper or in person. This method is more secure since spoofing techniques may be used to identify unauthorized users, and once someone has been authenticated, only they can access the system. Because it provides greater segmentation than other techniques, the ability of LBPH to capture these distinctions between genuine and spoofed faces has been employed as a method for separating features, although LBPH may not be adequate to identify all face spoofing attempts, this is why MobilenetV2 model is used to detect faces. MobileNetV2 is a promising deep-learning model for face spoofing detection due to its efficiency and effectiveness. However, to increase its accuracy and resilience in identifying various sorts of face spoofing attempts, it might need to be integrated with additional techniques like LBPH, texture analysis, or depth analysis. This architecture will be used for classification, and people may be categorized based on their student ID or employee ID. Even with a large dataset, this architecture can provide superior accuracy. Therefore, other techniques such as the MobileNetV2 model may also be used in combination with LBPH to improve the accuracy of face spoofing detection.

Keyphrases: Convolutional Neural Network (CNN), Enrollment process, Facial Obfuscation, Local Binary Pattern Histogram, MobileNet Version2, image recognition

@booklet{EasyChair:10033,
  author    = {Madhumitha Peruboina and M Ramesh and Venkatesh Jinka and Likitha Machapalli and Sravan Venkata Ganesh Badveli and V P M B Aarthi},
  title     = {Deep Learning Based Facial Obfuscation Using MobileNet},
  howpublished = {EasyChair Preprint 10033},
  year      = {EasyChair, 2023}}