Hand Segmentation Using PCA Principal Direction Analysis

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current image segmentation methods for gesture recognition, particularly in hand segmentation, face challenges in achieving high-quality images of the target object, leading to increased scanning range and error rates in fingertip extraction, which affects the accuracy of machine learning-based gesture recognition systems.

Innovation Solution

The method involves obtaining a connected region from a depth image using a depth camera, determining the principal direction through principal component analysis, and acquiring an image of the target object based on the relationship between the target object's morphology and the principal direction, thereby reducing error rates and enhancing image quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional image segmentation methods are used for hand segmentation, then the processing can be performed with simple algorithms, but the segmentation quality is poor leading to high error rates in fingertip extraction

Engineering Contradiction:

Improvefingertip extraction accuracyVSAvoidsegmentation process complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies segmentation by dividing the hand segmentation process into distinct stages: initial connected region extraction from depth image, principal direction determination through PCA, and final hand region extraction based on morphology-principal direction relationships. This multi-stage segmentation approach improves fingertip extraction accuracy while managing complexity through systematic decomposition of the segmentation task.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces principal direction analysis as an additional dimensional feature for hand segmentation. By determining the principal direction of the connected region and utilizing the relationship between target object morphology and principal direction, the method adds a new dimension of information beyond simple connected component analysis, thereby improving segmentation precision without proportionally increasing complexity.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Quantity of substance

If the scanning range is increased to capture more fingertip data, then more comprehensive gesture data can be obtained, but the error rate in fingertip extraction increases

Engineering Contradiction:

Improvefingertip data quantityVSAvoidfingertip extraction accuracy

Core Design Contradiction:

Quantity of substanceVSMeasurement precision

Solution Approach 1:

The patent applies local quality by determining the principal direction of the connected region and using the relationship between target object morphology and principal direction to guide the extraction process. This allows the system to focus processing quality on relevant regions (hand and fingertips) while maintaining comprehensive coverage, thereby increasing data quantity without proportionally increasing error rates.

Inventive Principle:

Principle #3Local quality

3Productivity

If simple connected region extraction is used from depth image, then the processing speed is fast, but the extracted region includes excessive background leading to poor image quality

Engineering Contradiction:

Improvesegmentation processing speedVSAvoidhand image quality

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent applies preliminary action by first extracting the connected region from the depth image to obtain an initial hand region, then performing principal direction determination on this pre-extracted region. This preliminary extraction step enables fast initial processing while subsequent refinement steps improve image quality, thereby maintaining high productivity while achieving good manufacturing precision.

Inventive Principle:

Principle #10Preliminary action

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach results in reduced scanning range and error rates for fingertip extraction, providing accurate test samples for machine learning-based gesture recognition systems by isolating the target object effectively from the depth image.

Implementation Method 1

obtaining a connected region where a target object is located from a depth image

Methodology Applied

Scientific EffectDepth sensing:

Implementation Method 2

determining a principal direction or a secondary direction of the connected region by a principal component analysis method

Methodology Applied

Scientific EffectPrincipal component analysis:

Data Source

PatentEP3537375B1Image segmentation methods, image segmentation system and device comprising same, and storage medium

Publication Date: 2021.09.01 BOE TECHNOLOGY GROUP CO LTD

Figure 1~4
Figure 5~9
Figure 10~13

AI summary

An image segmentation method, an image segmentation system, a storage medium and an apparatus including the image segmentation system or the storage medium are provided. The image segmentation method includes: obtaining a connected region where a target object is located from a depth image; determining a principal direction or a secondary direction of the connected region by a principal component analysis method; and acquiring an image of the target object from the connected region according to relationship between morphology of the target object and the principal direction or the secondary direction. Embodiments of the present disclosure acquire a high-quality image of the target object.