On-Device Face Authentication Using ROI Pooling to Reduce Computational Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based face authentication technologies require significant computational resources due to the extensive operations performed in face detectors and classifiers.
Innovation Solution
The proposed method minimizes computational load by using ROI pooling in the highest or multiple layers of the detection model to extract a fixed-length feature, which is then used for face classification without requiring a separate classification model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensive operations are performed in face detector and classifier using deep learning models, then authentication accuracy is improved, but computational load and system complexity increase significantly
Solution Approach 1:
The patent merges the face detection model and face classification model into a single integrated deep learning model. The detection model's feature extractor simultaneously performs both face detection and face classification functions, eliminating the need for separate classification models and reducing overall computational load while maintaining authentication accuracy.
Solution Approach 2:
The detection model is designed with multi-functionality, serving both as a face detector and a face classifier. The feature extractor within the detection model extracts features that are directly used for classification, making the single model universal for both detection and classification tasks, thereby reducing system complexity.
2Reliability
If separate face detection and classification models are used, then authentication performance is improved, but computational resources and processing time increase
Solution Approach 1:
The patent combines separate detection and classification models into one unified model structure. The feature extractor in the detection model produces features that are directly fed into the classification layer, creating an integrated pipeline that reduces redundant computations and lowers overall energy consumption while maintaining reliable authentication performance.
3Measurement precision
If traditional deep learning authentication methods are used, then authentication accuracy is achieved, but system efficiency decreases due to extensive operations
Solution Approach 1:
The patent integrates detection and classification operations into a single model execution, reducing the number of separate computational passes required. This merging of functions streamlines the authentication pipeline, improving system efficiency by eliminating redundant operations while preserving authentication accuracy through the shared feature extractor.
Data Source
AI summary
Disclosed are a method and apparatus for real-time on-device authentication based on deep learning. A deep learning-based authentication method includes detecting a location of a region of interest (ROI) occupied by a face portion an input image by using a detection model, extracting a feature map from the input image by using a feature extractor of the detection model, extracting a fixed length feature for the face portion using the feature map and ROI pooling for the detected location of the ROI, and classifying a face included in the input image based on the fixed length feature.