Compact CNN Model for Facial Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current convolutional neural networks (CNNs) for facial recognition are computationally complex, requiring large amounts of memory, processing power, and time, making them unsuitable for portable devices that need efficient and power-constrained facial recognition systems.
Innovation Solution
A compact CNN model with a set of five convolution layers and one fully connected layer is proposed, utilizing maximum feature map (MFM) operations and depth-wise separable convolution to reduce computational load, along with batch normalization to prevent overfitting, enabling efficient facial recognition and ocular detection with minimal computational overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a standard CNN model is used for facial recognition, then recognition accuracy is improved, but computational complexity and power consumption increase
Solution Approach 1:
The patent segments the CNN architecture into distinct functional blocks: convolutional layers for feature extraction, ReLU activation layers for non-linearity, and pooling layers for dimensionality reduction. This segmentation allows each component to be optimized independently for power efficiency while maintaining overall recognition accuracy.
Solution Approach 2:
The patent extracts and removes unnecessary computational components from the standard CNN model, retaining only the essential layers (convolutional, ReLU, pooling) required for accurate facial recognition. This extraction eliminates redundant computations that would increase power consumption on power-constrained devices.
2Measurement precision
If a standard CNN model is used for facial recognition, then recognition accuracy is improved, but device complexity increases
Solution Approach 1:
The patent segments the CNN into modular layers with clear functional distinctions. Each layer type (convolutional, ReLU, pooling) has a specific, simplified role, making the overall model structure easier to implement and manage on embedded devices while preserving recognition accuracy through proper layer sequencing.
Solution Approach 2:
The patent extracts only the necessary layers from a full CNN architecture, removing complex components such as fully connected layers and dropout layers that are not essential for the embedded facial recognition task. This extraction simplifies the model complexity while maintaining sufficient accuracy for the application.
3Measurement precision
If a standard CNN model is used for facial recognition, then recognition accuracy is improved, but processing time increases
Solution Approach 1:
The patent segments the processing pipeline into parallelizable stages: convolution operations, activation functions, and pooling operations. This segmentation enables efficient processing on embedded hardware by allowing certain operations to be executed in parallel or pipelined, reducing overall processing time while maintaining accuracy.
Solution Approach 2:
The patent extracts and removes computationally intensive layers from the standard CNN model, such as deep fully connected layers and complex normalization operations. By retaining only the essential convolutional, ReLU, and pooling layers, the model achieves acceptable accuracy with significantly reduced processing time suitable for real-time embedded applications.
Data Source
AI summary
Methods, systems, and devices for object recognition are described. Generally, the described techniques provide for a compact and efficient convolutional neural network (CNN) model for facial recognition. The proposed techniques relate to a light model with a set of layers of convolution and one fully connected layer for feature representation. A new building block of for each convolution layer is proposed. A maximum feature map (MFM) operation may be employed to reduce channels (e.g., by combining two or more channels via maximum feature selection within the channels). Depth-wise separable convolution may be employed for computation reduction (e.g., reduction of convolution computation). Batch normalization may be applied to normalize the output of the convolution layers and the fully connected layer (e.g., to prevent overfitting). The described techniques provide a compact and efficient CNN model which can be used for efficient and effective face recognition.