Face Recognition Multi-Scale LBP Feature Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing face recognition methods suffer from low recognition accuracy due to limitations in feature extraction and representation of face images.

Innovation Solution

The method involves down-sampling original face images to obtain images of different sizes, followed by block partition processing and feature extraction using Local Binary Pattern (LBP) descriptors, which increases the capability of face image features to describe the image and improves recognition accuracy by comparing similarity between image blocks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional single-scale LBP feature extraction is used, then the processing is simple, but the recognition accuracy is low

Engineering Contradiction:
Improverecognition accuracyVSAvoidfeature extraction complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The face image is divided into multiple scale levels through down-sampling, and each scale level is processed separately to extract LBP features. This segmentation allows the system to capture facial features at different resolutions, improving recognition accuracy while maintaining manageable processing complexity through hierarchical organization.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a new dimension of scale variation by generating down-sampled images at multiple resolution levels. This multi-scale approach adds a dimensional aspect to feature extraction, enabling the system to represent facial characteristics across different sizes and improve overall recognition performance.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Adaptability or versatility

If single-size face image features are extracted, then the feature extraction is straightforward, but the feature description capability is insufficient

Engineering Contradiction:
Improvefeature description capabilityVSAvoidfeature extraction process
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The multi-scale LBP feature extraction system serves multiple functions simultaneously: it extracts features at different resolutions, captures both local and global facial characteristics, and provides robust representation across varying image sizes. This multi-functionality enhances the versatility of feature description without requiring separate processing systems.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If block partition processing is applied to original and down-sampled images, then recognition accuracy improves, but processing time increases

Engineering Contradiction:
Improverecognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs down-sampling and block partitioning as preliminary actions before feature extraction. By pre-processing the images into appropriate blocks and scales, the subsequent LBP feature extraction becomes more efficient, and the overall processing time is managed through structured preparation steps that facilitate faster recognition.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP2835762B1Face recognition method and device
Publication Date: 2018.06.27 HUAWEI TECH CO LTD
  • EP2835762B1 patent drawingFigure 1
  • EP2835762B1 patent drawingFigure 2
  • EP2835762B1 patent drawingFigure 3~5

AI summary

The present invention provides a face recognition method and device. The method includes: performing down-sampling of different sizes on an original face image; performing block partition processing on a down-sampled image and the original face image; extracting feature of each image block obtained after block partition processing is performed; performing similarity matching between the extracted feature of each image block and features of image blocks corresponding to a pre-registered face image to obtain a similarity of each image block; and obtaining a face image recognition result according to an obtained similarity between all the image blocks. According to the face recognition method and device I provided in the present invention, down-sampling of different sizes is performed on an original face image; features of face images of multiple sizes are extracted, which increases a capability of face image features to describe the face image; block partition processing is performed on a down-sampled image and the original face image, and a face image recognition result is obtained according to a similarity between all the obtained image blocks, which increases an accuracy of face image recognition.