Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
97 tokens/sec
GPT-4o
53 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
5 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Disguised Face Identification (DFI) with Facial KeyPoints using Spatial Fusion Convolutional Network (1708.09317v1)

Published 30 Aug 2017 in cs.CV

Abstract: Disguised face identification (DFI) is an extremely challenging problem due to the numerous variations that can be introduced using different disguises. This paper introduces a deep learning framework to first detect 14 facial key-points which are then utilized to perform disguised face identification. Since the training of deep learning architectures relies on large annotated datasets, two annotated facial key-points datasets are introduced. The effectiveness of the facial keypoint detection framework is presented for each keypoint. The superiority of the key-point detection framework is also demonstrated by a comparison with other deep networks. The effectiveness of classification performance is also demonstrated by comparison with the state-of-the-art face disguise classification methods.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Amarjot Singh (9 papers)
  2. Devendra Patil (2 papers)
  3. G Meghana Reddy (1 paper)
  4. SN Omkar (2 papers)
Citations (52)

Summary

  • The paper introduces a novel Spatial Fusion CNN framework that uses 14 facial keypoints in a star-net configuration to enhance recognition of disguised faces.
  • It employs two newly curated datasets, each with 2000 images, to train the network from scratch and minimize reliance on transfer learning.
  • Empirical results demonstrate significant improvements in keypoint detection and classification accuracy over existing state-of-the-art methods.

Disguised Face Identification with Facial KeyPoints Using Spatial Fusion Convolutional Network

The paper introduces a novel framework for Disguised Face Identification (DFI) that leverages deep learning techniques to tackle the complex problem of recognizing faces with disguises. The authors propose using facial key-points for enhanced face identification accuracy, an approach that addresses the challenges posed by face recognition systems when faced with alterations such as wigs, eyeglasses, and various other disguise elements. The paper also introduces two new datasets, designed to support and improve the training of deep convolutional networks specifically for the task of disguised face identification.

Framework Overview

Central to the proposed methodology is the Spatial Fusion Convolutional Network, which is tasked with detecting 14 critical facial key-points. These key-points are further utilized to form a star-net structure, which significantly aids in the classification phase of face identification. The framework promises increased extraction performance compared to existing systems, as demonstrated by the results presented in the paper.

Contributions and Datasets

The primary contributions of the paper lie in the development of the Spatial Fusion Network-based DFI framework and the introduction of new annotated datasets specifically curated for facial disguise challenges. The datasets, which comprise simple and complex backgrounds, provide 2000 images each with varied disguise combinations. These datasets are strategically aimed at reducing the dependency on transfer learning, moving towards training from scratch with a more applicable range of training data.

Key-Point Detection Performance

The paper presents an extensive evaluation of key-point detection, showcasing noteworthy accuracy improvements. The accuracy of key-point detection was meticulously recorded for each of the 14 key-points, with figures indicating varying accuracy levels depending on the complexity of the background (simple vs. complex). The results underscore the efficacy of the Spatial Fusion Convolutional Network, which achieved superior performance relative to other network architectures.

Classification and Comparison

In terms of classification, the proposed DFI framework's efficacy is underscored by comparative analyses with state-of-the-art methodologies, where it demonstrates a marked improvement in accuracy rates. This is attributed to the robust star-net structure and the orientation-based classification approach utilized, which has shown resilience against diverse disguises under various conditions.

Implications and Future Work

While the results presented indicate strong potential for practical applications, including security and surveillance, the primary implication of this work extends into theoretical domains, offering a foundational approach for future AI and machine learning models tackling disguised and altered appearances in facial identification tasks. Future work could extend to evaluating the framework against larger, more varied datasets or incorporating more advanced forms of convolutional architectures to further optimize both detection and classification tasks.

In summary, the paper provides a significant addition to the field of computer vision, especially in the context of face recognition under disguise. By addressing the data scarcity issue with new datasets and leveraging spatial fusion techniques, the framework opens pathways for subsequent innovations in robust facial recognition technologies.

Youtube Logo Streamline Icon: https://streamlinehq.com