Real-Time Periocular Based Face Recognition System

Abstract

This invention presents a real-time periocular-based face recognition system that provides robust feature extraction and accurate face detection under varying conditions, including lighting, occlusions, and crowded settings. Utilizing edge detection, Gabor filters, and Local Binary Patterns (LBP), the system captures periocular features effectively, supported by adaptive illumination normalization and occlusion-handling techniques. With applications in biometric authentication and surveillance, this framework is designed to maintain high accuracy across diverse scenarios, providing a scalable and reliable solution for real-world security needs.

Specification

Description:Field of the Invention

[01] This invention relates to biometric systems, specifically a real-time periocular-based face recognition system focused on periocular feature extraction. It aims to improve identification accuracy under challenging conditions, including lighting variations, occlusions, and crowded environments, and provides a robust framework for biometric applications in security and surveillance.

Description of Related Art

[02] Traditional face recognition approaches struggle with variable conditions such as lighting, occlusions, and accessories, making accurate feature extraction challenging. Previous solutions utilizing edge detection, Gabor filters, and Local Binary Patterns (LBP) perform adequately in controlled settings but lack adaptability for real-world applications with unpredictable lighting and occlusions.

[03] Existing systems often face limitations in detecting multiple faces in crowded settings and handling varied facial orientations, which restricts their usability for surveillance and crowd control. These systems lack integration of robust techniques for dynamic conditions, necessitating a framework with advanced feature extraction and adaptability.

[04] Consequently, there is a need for a robust, adaptive biometric system that excels in extracting periocular features under diverse conditions and offers accurate face detection in complex, crowded environments. The present invention aims to address these limitations through an integrated, periocular-based recognition approach.

SUMMARY

[05] The invention introduces a real-time, periocular-based face recognition system that incorporates edge detection, Gabor filters, and Local Binary Patterns (LBP) to extract unique features from the periocular region. This system enhances the stability and accuracy of biometric recognition in variable lighting and occlusion conditions.

[06] The system leverages adaptive illumination normalization and occlusion handling techniques to enhance feature extraction accuracy. It includes an integrated face identification module optimized for detecting and counting faces in crowded settings, which supports operational efficiency in surveillance.

[07] A significant aspect of the invention is its ability to align periocular features across different viewpoints through multi-view matching, providing robustness against profile variations. This capability enhances the effectiveness of periocular matching in real-world applications.

[08] Additionally, the system handles accessory-induced occlusions by employing image inpainting and adversarial training techniques, allowing reliable recognition despite the presence of glasses, masks, or other obstructive elements.

[09] The invention thus offers an advanced, real-time periocular-based biometric system designed for security and surveillance applications, capable of accurate recognition in challenging environments.

[0010] This innovative framework integrates modules for adaptive illumination normalization, multi-view matching, and occlusion handling, providing a comprehensive solution for real-time periocular-based recognition across various conditions.

BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The following provides an overview of the features and advantages of embodiments of the present invention. The detailed description is accompanied by drawings that demonstrate system functionality:
[0012] The invention illustrates the primary components of the periocular-based recognition system, including feature extraction and face detection modules, outlining the data flow from image acquisition to feature analysis.

[0013] The invention depicts adaptive illumination normalization and occlusion-handling techniques, visualizing the process of feature extraction under varied lighting and obstruction conditions.

DETAILED DESCRIPTION

[0014] The invention utilizes periocular feature extraction techniques, including edge detection, Gabor filters, and Local Binary Patterns (LBP), to capture distinct features around the eye region. This methodology enables robust identification under variable lighting, occlusions, and crowded settings, making it highly suitable for biometric applications.

[0015] Modules
• Data Collection: A dataset of periocular images under various conditions, including different lighting and occlusions, is gathered for effective model training.
• Preprocessing: Standardizes images by resizing and converting them to grayscale. Histogram equalization is applied for improved contrast, enhancing the robustness of feature extraction.
• Feature Extraction: Utilizes edge detection for defining periocular boundaries, Gabor filters for capturing texture, and LBP for detailed local texture patterns, ensuring consistent feature extraction.
• Adaptive Illumination Compensation: Includes histogram equalization and image normalization to address lighting variations, enabling accurate feature extraction across different environments.
• Multi-View Periocular Matching: Employs affine transformations and geometric alignment to match periocular features from varying angles, improving recognition across profile changes.
• Accessory and Occlusion Handling: Incorporates image in painting, adversarial training, and feature augmentation to mitigate the effects of accessories, making the system resilient to obstructions like glasses and masks.
• Crowded Environment Detection and Multi-Face Recognition: Utilizes deep learning-based object detection (e.g., Faster YOLOV3) for detecting and counting multiple faces in congested environments, with clustering and template matching for accurate face identification and crowd control.

[0016] This integrated approach ensures high reliability in face recognition, providing a system suitable for real-world applications in surveillance and security.

[0017] The periocular-based face recognition system incorporates a face counting mechanism that enables accurate person count and recognition in crowded scenes, enhancing security applications in public spaces.
[0018] Edge detection, Gabor filters, and LBP form the core of feature extraction, while adaptive techniques for illumination and occlusion handling further strengthen system robustness, addressing challenges in traditional systems.
[0019] The system architecture supports modular integration, enabling it to adapt to various hardware platforms and environments, ensuring scalability for large-scale deployment.
[0020] Overall, the invention offers a robust, real-time solution for periocular-based recognition, optimized for challenging conditions in biometric authentication and surveillance applications.
, Claims:I/We Claim:

1. A Real-Time Periocular-Based Face Recognition System comprising:
- A periocular feature extraction module utilizing edge detection, Gabor filters, and Local Binary Patterns (LBP) to capture periocular features reliably under variable lighting conditions.
2. The system of claim 1, wherein adaptive illumination normalization methods are applied, such as histogram equalization, to maintain feature extraction consistency across different lighting environments.
3. The system of claim 1 includes a multi-view periocular matching module, which uses geometric transformations to align periocular features from various angles, ensuring accurate recognition across profile variations.
4. The system of claim 1 includes a face identification component that utilizes clustering and template matching to detect and count multiple faces in crowded settings.
5. The system of claim 1 further includes an occlusion-handling module, incorporating image in painting and adversarial training to address obstructions like glasses and masks for consistent periocular recognition.
6. The system of claim 1, wherein PCA is utilized in preprocessing, to enhance dimensionality reduction and improve computational efficiency in periocular feature extraction.
7. A system, method, and computer-readable medium for periocular-based face recognition, which integrates feature extraction, multi-view matching, and adaptive illumination normalization, ensuring reliable biometric recognition under challenging real-world conditions.

Documents