Face Recognition Neural Network Segmentation for Feature Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional face recognition devices require multiple forward computations for feature extraction from generated patches, leading to long processing times and high resource consumption, and lack reliability in feature averaging, making them inefficient and unreliable.

Innovation Solution

A face-recognizing device that uses a feature extraction network to generate multiple features from a single input image through convolution and pooling operations, followed by a neural aggregation network to combine these features using quality scores, thereby producing an optimal feature for recognition without patch generation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple patches are generated and processed through forward computation to extract features, then feature extraction completeness is improved, but processing time and computing resource consumption increase significantly

Engineering Contradiction:
Improvefeature extraction completenessVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the feature extraction process into two independent networks: a feature extraction network that processes the original image once to generate feature maps, and a patch generation network that generates multiple patches from these feature maps. This segmentation allows the system to obtain features from multiple patches without performing forward computation multiple times on the original image, thus reducing processing time while maintaining feature extraction completeness.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary feature extraction on the original image before generating patches. The feature extraction network processes the original image once to create feature maps, which are then used to generate multiple patches. This preliminary action eliminates the need for repeated forward computations on the original image for each patch, significantly reducing processing time and computational resource consumption.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If multiple patches are generated and processed to extract features, then feature extraction completeness is improved, but computing resource consumption increases significantly

Engineering Contradiction:
Improvefeature extraction completenessVSAvoidcomputing resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent divides the computing task into two stages: first, the feature extraction network processes the original image once to generate feature maps; second, the patch generation network creates multiple patches from these feature maps. This segmentation avoids repeated full forward computations, reducing computing resource consumption while maintaining comprehensive feature extraction.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary feature extraction on the original image before generating patches. By extracting features once and then generating multiple patches from these pre-extracted features, the system avoids redundant computing operations, significantly reducing energy and computing resource consumption.

Inventive Principle:
Principle #10Preliminary action

3Productivity

If features from multiple patches are averaged to perform face recognition, then recognition is performed, but reliability of recognition results cannot be ensured

Engineering Contradiction:
Improverecognition performanceVSAvoidrecognition result reliability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent introduces a feedback mechanism where the patch generation network generates multiple patches from the same feature maps, and the system selects or combines patches based on their quality or relevance. This feedback-based approach ensures that the most reliable patches are used for face recognition, improving the reliability of recognition results compared to simple averaging of all patches.

Inventive Principle:
Principle #23Feedback

4Reliability

If conventional face recognition devices use input augmentation to improve facial recognition performance, then recognition performance is improved, but processing time and resource consumption increase

Engineering Contradiction:
Improvefacial recognition performanceVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent segments the processing into feature extraction and patch generation stages, allowing input augmentation (multiple patches) to be achieved without proportionally increasing processing time. The feature extraction network processes the original image once, and the patch generation network creates multiple augmented views from these features, improving recognition performance without linearly increasing processing time.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3690721B1Method for recognizing face using based on deep neural network
Publication Date: 2026.02.25 STRADVISION
  • EP3690721B1 patent drawingFigure 1
  • EP3690721B1 patent drawingFigure 2
  • EP3690721B1 patent drawingFigure 3

AI summary

A method for face recognition by using a multiple patch combination based on a deep neural network is provided. The method includes steps of: a face-recognizing device, (a) if a face image with a 1-st size is acquired, inputting the face image into a feature extraction network, to allow the feature extraction network to generate a feature map by applying convolution operation to the face image with the 1-st size, and to generate multiple features by applying sliding-pooling operation to the feature map, wherein the feature extraction network has been learned to extract a feature using a face image for training having a 2-nd size and wherein the 2-nd size is smaller than the 1-st size; and (b) inputting the multiple features into a learned neural aggregation network, to allow the neural aggregation network to aggregate the multiple features and to output an optimal feature for the face recognition.