Logo image
Is CNN Really Looking at Your Face?
Conference paper   Peer reviewed

Is CNN Really Looking at Your Face?

Hiroya Kawai, Takashi Kozu, Koichi Ito, Hwann-Tzong Chen and Takafumi Aoki
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol.13188 LNCS, pp.525-539
2022

Abstract

Biometrics CNN Face parsing Face recognition Theoretical Computer Science Computer Science (all)
Face recognition has dramatically improved its performance with the advent of deep learning, especially Convolutional Neural Networks (CNNs), while they have raised a new issue: the difficulty in interpreting the results. The question is what CNN looks at in a face image to identify a person. To answer this question, this paper presents a simple and novel analysis of deep face recognition based on facial parts. We evaluate the recognition accuracy of face images with specific regions masked using face segmentation labels. Our analysis clarifies what CNNs really need in face images for face recognition. The paper concludes with an application of face recognition models to general visualization methods and the problems contained in some classical face image datasets.

Metrics

1 Record Views

Details

Logo image