CLASSIFICATION OF DETECTION AND RECOGNITION METHODS OF THE PERSON ON THE IMAGE

Dmitry Kuznetsov, P.g. Nikolsky, Dmytro S. Rachkov, Andrey Kuznetsov, A.p. KhakhamovResearch 2019 · DOI

Abstract

This paper provides a comprehensive taxonomy of face detection and recognition algorithms, categorizing them based on their underlying logic and mathematical approaches. The authors evaluate these methods to identify the most effective combination for a "smart meeting room" system, ultimately recommending the Viola-Jones algorithm for detection and Active Appearance Models for recognition.

Introduction

As "intellectual halls" (smart meeting rooms) become more common, there is a growing need for seamless, automated access control and participant accounting. While facial recognition is the obvious solution, the sheer volume of available algorithms—each with different trade-offs in speed, accuracy, and computational cost—makes it difficult for engineers to choose the right tools. The authors address this by creating a structured classification of the field, helping researchers navigate the landscape of empirical vs. modeling-based detection and local vs. global recognition features.

Face Detection Taxonomy

Face detection is the mandatory first step in any facial analysis pipeline; recognition cannot happen without an accurate bounding box. The authors divide detection methods into two primary categories: Empirical methods and Image Modeling.

Taxonomy of face detection methods

Empirical methods include template matching and top-down/bottom-up approaches. Image modeling is further subdivided into non-adaptive methods (like Principal Component Analysis) and adaptive methods. Among these, the Viola-Jones algorithm is highlighted as the best choice for real-time search in video streams due to its efficiency using Haar-like features and cascading classifiers.

Face Recognition Taxonomy

Recognition methods are categorized by the type of features used: local features and global features. This distinction separates methods that model the relationship between specific landmarks (like eyes, nose, and mouth) from those that analyze the face as a whole.

Taxonomy of face recognition methods
  • Local Features: Includes graph matching, Hidden Markov Models, and informative regions.
  • Global Features: Includes discriminant analysis like Linear Discriminant Analysis (LDA) and Principal Component Analysis (PCA), as well as various neural network architectures.

Key Insights and Recommendations

A key insight from the study is the power of hybrid modeling. The authors highlight Active Appearance Models (AAM) as particularly effective because they combine shape constraints with texture information, making them robust for identifying individuals in dynamic environments where facial expressions and lighting may vary.

For practical implementation in a smart meeting room, the authors recommend the following stack:

  1. Viola-Jones for high-speed, real-time detection.
  2. Active Appearance Models (AAM) for robust recognition.

Conclusion

While deep learning (Neural Networks) offers significant potential for modeling complex data distributions, the computational efficiency of the Viola-Jones and AAM combination provides a more practical solution for real-time applications in current smart room infrastructures. Future work will involve the practical implementation and validation of this theoretical framework.

Summary updated

Abstract

. , , . , . , , . . .
BibTeX
@inproceedings{kuznetsov2019classification,
  author = {Dmitry Kuznetsov and P.G. Nikolsky and Dmytro S. Rachkov and Andrey Kuznetsov and A.P. Khakhamov and A.P. Khakhamov},
  title = {{CLASSIFICATION OF DETECTION AND RECOGNITION METHODS OF THE PERSON ON THE IMAGE}},
  booktitle = {Research},
  year = {2019},
  doi = {10.18413/2518-1092-2019-4-1-0-6},
}
Figure