Description:FIELD OF THE INVENTION
The invention relates to biometric security and, more particularly, to systems and methods for detecting facial spoofing attacks. It concerns a hybrid deep learning architecture using convolutional neural networks (CNN) for feature extraction and generative adversarial networks (GAN) for adversarial training to provide robust, real-time detection of fake faces in digital authentication systems.
BACKGROUND OF THE INVENTION
Especially facial recognition technologies, biometric systems have become rather well-known in the modern digital era for purposes in security, surveillance, banking, and personal authentication. But the spread of advanced picture and video editing tools as well as the creation of deepfake technologies have made these systems progressively prone to spoofing attacks vulnerable. Such attacks could fool authentication systems using synthetic produced faces, video footage, or printed images. Particularly those generated with deep learning-based generative models, current methods sometimes fail to find high-quality spoofing media. Thus, a sophisticated solution that can regularly tell real-time between genuine and fake faces is absolutely needed.
US2011254942A1: A masquerading detection system includes: an imaging unit (2) that obtains a first image by imaging an inspection object (12) from a first angle, and a second image by imaging the inspection object from a second angle which is different from the first angle; a unit (101) that detects first feature points from the first image, obtains first feature point coordinates of the detected feature points, detects second feature points from the second image, and obtains second feature point coordinates of the detected feature points; a unit (104) that obtains transformed coordinates by performing a plane projective transformation for the second feature point coordinates from the second image to the first image; and a unit (105) that determines that masquerading has been attempted when the difference between the transformed coordinates and the corresponding first feature point coordinates is equal to or smaller than a predetermined value.
US10987030B2: A computer implemented method for detecting an attempt to spoof and facial recognitions apparatus determines for a plurality of spatially separated regions of a surface, a respective measure of at least one vital sign. A determination is made from the respective measures of at least one vital sign, homogeneity information associated with the respective measures, the homogeneity information is used to determine if said spatially separate regions of said surface are living tissue.
Traditional facial recognition systems are vulnerable to spoofing using printed photos, replayed videos, or AI-generated deepfakes. Existing anti-spoofing methods—texture analysis, motion analysis, hardware depth sensors—often have high false positives, limited spoof-type coverage, or require special equipment. They struggle to detect high-quality synthetic faces generated by modern deep learning techniques. The present invention solves these shortcomings by combining CNN-based micro-texture analysis with GAN-based spoof simulation and adversarial training, resulting in a system that is accurate, adaptive, cost-effective, and deployable on standard cameras without special hardware.
SUMMARY OF THE INVENTION
This summary is provided to introduce a selection of concepts, in a simplified format, that are further described in the detailed description of the invention.
This summary is neither intended to identify key or essential inventive concepts of the invention and nor is it intended for determining the scope of the invention.
The invention provides a hybrid facial spoofing detection system integrating convolutional neural networks (CNNs) and generative adversarial networks (GANs). A camera acquires facial images or video streams and passes them through a preprocessing module for alignment, normalisation, and noise reduction. A CNN extracts fine-grained facial features such as skin texture, reflectance, and micro-expressions. Concurrently, a GAN generates synthetic spoof faces and trains a discriminator to detect subtle artefacts. This adversarial learning feedback strengthens the CNN’s ability to classify real versus fake faces.
The system architecture includes modules for image acquisition, preprocessing, CNN-based feature extraction, GAN-based spoof simulation and adversarial training, classification, and output integration. The classification module assigns a binary label (REAL/FAKE) with a confidence score. Outputs can be integrated into existing biometric systems, triggering access decisions or alerts.
By eliminating dependency on specialised sensors and adapting automatically to new attack types, the invention achieves high detection accuracy and real-time performance on common mobile or webcam devices. It is suited for banking, border security, eKYC, and other critical applications.
To further clarify advantages and features of the present invention, a more particular description of the invention will be rendered by reference to specific embodiments thereof, which is illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. The invention will be described and explained with additional specificity and detail with the accompanying drawings.
The proposed invention introduces Generative Adversarial Networks (GANs) for adversarial learning and robust classification and Convolutional Neural Networks (CNNs) for feature extraction, a hybrid facial spoofing detection system is proposed.
BRIEF DESCRIPTION OF THE DRAWINGS
The illustrated embodiments of the subject matter will be understood by reference to the drawings, wherein like parts are designated by like numerals throughout. The following description is intended only by way of example, and simply illustrates certain selected embodiments of devices, systems, and methods that are consistent with the subject matter as claimed herein, wherein:
Fig1. Overall System Architecture
Figure 2: Overall Architecture of Facial Spoofing Detection System
Figure 3: GAN-CNN Integration in Facial Spoofing Detection
The figures depict embodiments of the present subject matter for the purposes of illustration only. A person skilled in the art will easily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.
DETAILED DESCRIPTION OF THE INVENTION
The detailed description of various exemplary embodiments of the disclosure is described herein with reference to the accompanying drawings. It should be noted that the embodiments are described herein in such details as to clearly communicate the disclosure. However, the amount of details provided herein is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the scope of the present disclosure as defined by the appended claims.
It is also to be understood that various arrangements may be devised that, although not explicitly described or shown herein, embody the principles of the present disclosure. Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples, are intended to encompass equivalents thereof.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a",” “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises,” “comprising,” “includes” and/or “including,” when used herein, specify the presence of stated features, integers, steps, operations, elements and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and/or groups thereof.
It should also be noted that in some alternative implementations, the functions/acts noted may occur out of the order noted in the figures. For example, two figures shown in succession may, in fact, be executed concurrently or may sometimes be executed in the reverse order, depending upon the functionality/acts involved.
In addition, the descriptions of "first", "second", “third”, and the like in the present invention are used for the purpose of description only, and are not to be construed as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Thus, features defining "first" and "second" may include at least one of the features, either explicitly or implicitly.
Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which example embodiments belong. It will be further understood that terms, e.g., those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The proposed invention introduces Generative Adversarial Networks (GANs) for adversarial learning and robust classification and Convolutional Neural Networks (CNNs) for feature extraction, a hybrid facial spoofing detection system is proposed.
Mostly depending on simple image matching or feature-based detection, conventional facial authentication systems are insufficient against high-quality spoofing attacks. Combining Generative Adversarial Networks (GAN) with Convolutional Neural Networks (CNN) helps our system to raise detection capacity against both known and hitherto unmet spoofing techniques.
• CNN feature extraction detects minute textures and patterns in facial images separating real skin from printed or digital artifacts.
• GAN-based synthetic face generation lets one mimic spoof attacks. The discriminator network of GAN is trained to identify small anomalies in synthetic images, so enhancing CNN's ability to distinguish real from phoney faces.
System Implementation:
The proposed invention includes the following modules:
1. Image Acquisition Module
 Captures live facial data through webcams or mobile phone cameras.
 Supports both still images and video streams.
2. Preprocessing Module
 Performs normalization, cropping, and face alignment.
 Removes background noise and enhances facial features.
3. CNN-Based Feature Extraction
 Uses a lightweight convolutional model to extract multi-scale facial features.
 Focuses on micro-texture analysis (e.g., skin reflectance, edge distortions, eye blinking).
4. GAN-Based Spoof Simulation and Adversarial Training
 The generator creates fake images from noise.
 The discriminator learns to differentiate between real and generated (fake) images.
 These learnings are fed into the CNN to improve its robustness against novel spoof attacks.
5. Classification Module
 Takes output from CNN and assigns a label: REAL or FAKE.
 Provides a confidence score for decision support.
6. Output and Integration
 Outputs can be integrated into existing biometric systems or APIs.
 Supports alert systems in case of spoof detection (e.g., deny access, trigger alarm).
Workflow Overview
1. A user shows the camera their face.
2. The system corrects and preprocesses the picture.
3. CNN generates important facial features (e.g., edge sharpness, skin texture).
4. GAN refines the discriminator and creates fake face samples concurrently.
5. CNN gains detection accuracy by means of adversarial feedback, so strengthening it.
6. The system highly confidentially marks the input as REAL or FAKE.
7. The outcome is shown or included into a choice on authentication.
This multi-stage hybrid approach provides:
 High accuracy (92–96%)
 Robustness to emerging deepfake methods
 Real-time execution capability
 Ease of integration into mobile, cloud, and enterprise security systems
A deep learning-based facial spoofing detection system that combines convolutional neural networks (CNNs) and generative adversarial networks (GANs) is depicted in its "Overall System Architecture" diagram. The Input Image module is where the process starts. Here, a camera captures a face image, which could be a real human face or a spoof image (like a photo, video, or AI-generated deepfake). This picture is sent to the following stage for analysis and acts as the system's input.
After that, necessary tasks to standardize the input image are handled by the Preprocessing module. Face detection, alignment, resizing, grayscale conversion, and noise reduction are a few examples of these processes. By guaranteeing that the image is in a consistent format, preprocessing enhances the model's performance and dependability during classification.
The GAN-CNN Model, which consists of two main sub-modules, is the central component of the system. A discriminator and a generator make up a Generative Adversarial Network (GAN). While the discriminator learns to distinguish between a generated and real input face, the generator simulates a wide range of attack types by producing synthetic spoof face images from random noise. The model's resilience to fresh and complex spoofing attempts is improved by this adversarial training. Concurrently, the input face's texture, reflectance, depth inconsistencies, and micro-expressions are extracted by the Convolutional Neural Network (CNN). These characteristics are essential for spotting the subtle clues that set real faces apart from fakes.
The information is sent to the Classification module after the spoof simulation and feature extraction are complete. This part determines if the face is real or fake by analyzing the output feature map. Usually, a softmax layer or fully connected neural network is used to give the image a binary class. The Output Module, which prominently marks the input as either a Real Face or a Fake Face, displays the outcome at the end. Access to real-time systems, including banking interfaces, security gateways, mobile apps, and identity verification platforms, can then be granted or denied using this output.
By incorporating deep feature extraction and adversarial learning, this architecture successfully mitigates the weakness of conventional face recognition systems. GAN and CNN work together to give the system high accuracy, scalability, and real-time deployment capabilities while also learning from current spoof attacks and adapting to new, unseen ones.
The system operates in the following steps:
1. Data Collection: Images and videos are captured via camera or mobile interface.
2. Preprocessing: Input is normalized and augmented to enhance learning.
3. Feature Extraction (CNN): Facial features such as skin texture, depth, and edge consistency are extracted.
4. Adversarial Training (GAN): GANs generate synthetic spoof images to train the discriminator model.
5. Spoof Detection Module: The model identifies whether the input is real or fake based on learned features.
6. Output Layer: A binary classification label (Real / Fake) is returned with confidence score.
This architecture can be deployed on cloud infrastructure or edge devices to support real-time applications in mobile banking, security checkpoints, and eKYC systems.
The novelty of the proposed invention is several. First it shows a hybrid architecture greatly enhancing detection robustness by combining CNN-based feature analysis with GAN-generated spoof synthesis. Second, since the system does not depend on hardware-based liveness detection mechanisms, such depth-sensing or infrared cameras, it is cost-effective and deployable on common mobile or webcam devices. Third, the system consists of an adaptive learning mechanism whereby fresh synthetic spoof images generated by the GAN periodically re-training the CNN. This keeps the model current against developing issues including 3D-masked attacks and deepfake evolution. Furthermore supporting real-time inference, the architecture helps to ensure perfect integration into time-sensitive security systems. At last, the optional explainability module which emphasizes questionable facial areas influencing the classification offers transparency, a quality lacking in most traditional systems.
The invention consists of an image acquisition module capturing live facial data through standard cameras, supporting both still images and video streams.
A preprocessing module standardises input images by performing cropping, alignment, resizing, grayscale conversion, and noise reduction. This enhances the reliability and consistency of downstream feature extraction.
A convolutional neural network (CNN) module extracts multi-scale facial features such as micro-texture, edge sharpness, skin reflectance, and blinking patterns. These features help differentiate genuine human faces from printed or digital artefacts.
A generative adversarial network (GAN) module comprises a generator and a discriminator. The generator synthesises spoof images from noise to mimic various attack types. The discriminator learns to distinguish real from generated faces. This adversarial learning creates a rich training environment for the CNN.
The CNN benefits from the GAN’s discriminator outputs, improving robustness against both known and unseen spoofing techniques. By continuously generating new synthetic attacks, the system adapts to evolving threats such as 3D masks or high-resolution deepfakes.
A classification module receives the combined features and outputs a binary decision—REAL or FAKE—with a confidence score. This decision can be used directly in access control or fed to an external security system.
An output and integration module displays the classification result and communicates with external systems via APIs to grant or deny access, log incidents, or trigger alarms.
The system may optionally include an explainability module highlighting facial regions that influenced the classification decision, providing transparency to operators.
The hybrid architecture is lightweight, enabling deployment on edge devices like mobile phones or embedded systems. It can also run on cloud infrastructure to support large-scale applications.
Adversarial training is performed periodically, using newly generated synthetic faces to refresh the CNN’s parameters. This continual learning maintains high performance against new spoof types without manual intervention.
Latency is minimised by optimising the CNN architecture for real-time inference and by using efficient data pipelines between modules.
Security and privacy of captured facial data are maintained through encryption, secure storage, and controlled access to training datasets.
The system does not require specialized hardware such as depth sensors or infrared cameras, making it cost-effective and widely deployable.
By combining deep feature extraction and adversarial learning, the invention significantly outperforms conventional anti-spoofing methods in accuracy, adaptability, and speed.
This design is scalable, robust, and integrates seamlessly into existing authentication systems, enhancing their resilience to sophisticated spoofing attacks.
In the preferred embodiment, facial images are captured via a standard webcam or mobile camera and preprocessed to align and normalize faces. The CNN module extracts multi-scale facial features, while the GAN module generates synthetic spoof attacks and trains a discriminator. The CNN model is updated with adversarial feedback from the GAN to strengthen classification. The classification module outputs a REAL/FAKE label with a confidence score. Integration with an authentication system allows real-time access decisions. This configuration provides high detection accuracy (over 90%), low latency, and adaptability without requiring special hardware.
, Claims:1. A facial spoofing detection system comprising:
an image acquisition module configured to capture facial images or video streams from a standard camera;
a preprocessing module configured to normalise and enhance the captured images;
a convolutional neural network (CNN) module configured to extract multi-scale facial features;
a generative adversarial network (GAN) module comprising a generator configured to produce synthetic spoof images and a discriminator configured to distinguish real from generated images;
a classification module configured to determine whether the input face is real or fake based on the extracted features and adversarial feedback; and
an output integration module configured to communicate the classification result to external systems for authentication decisions.
2. The system as claimed in claim 1, wherein the preprocessing module performs face detection, alignment, resizing, grayscale conversion, and noise reduction.
3. The system as claimed in claim 1, wherein the CNN module analyses micro-texture, skin reflectance, edge sharpness, and blinking patterns to differentiate real faces from spoofed images.
4. The system as claimed in claim 1, wherein the GAN module continuously generates new synthetic spoof images to update the CNN module for robustness against evolving attack types.
5. The system as claimed in claim 1, wherein the output integration module provides a binary classification label with a confidence score and interfaces with external biometric systems via an application programming interface.
6. A method for detecting facial spoofing comprising:
capturing a facial image or video stream using a standard camera;
preprocessing the captured image to normalise and enhance facial features;
extracting multi-scale facial features using a convolutional neural network;
generating synthetic spoof images using a generative adversarial network and training a discriminator to distinguish real from generated images;
adversarially updating the convolutional neural network with feedback from the discriminator; and
classifying the input face as real or fake and communicating the result to an external authentication system.
7. The method as claimed in claim 6, wherein the preprocessing includes face detection, alignment, resizing, grayscale conversion, and noise reduction.
8. The method as claimed in claim 6, wherein the convolutional neural network analyses micro-texture, skin reflectance, edge sharpness, and blinking patterns to detect spoofing.
9. The method as claimed in claim 6, wherein synthetic spoof images are periodically generated to retrain the convolutional neural network for new attack types.
10. The method as claimed in claim 6, wherein the classification result includes a binary label and a confidence score integrated into an external biometric authentication system.