3D Image Segmentation via Sparse Local Feature Graphs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for non-linear depth adjustment in 3D image segmentation require significant computational resources, making them inefficient for real-time processing on devices with limited capabilities, such as mobile devices.

Innovation Solution

The method employs a sparse graph-based approach for 3D image segmentation, where local features are determined, a local feature graph is constructed, and edge weights are assigned, allowing for efficient segmentation by quantizing depth maps, identifying reliable texture regions, and computing histograms to segment the image into depth regions, thereby reducing computational load.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional pixel-based methods are used for 3D image segmentation, then segmentation accuracy is improved, but computational resource consumption increases significantly

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent segments the image into superpixels first, then performs graph construction and segmentation only on these superpixels rather than individual pixels. This hierarchical segmentation approach maintains segmentation accuracy while dramatically reducing the number of elements processed in subsequent computational steps, thus lowering computational resource consumption.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts and removes unreliable texture regions from further processing by calculating confidence values and eliminating regions below a threshold. This extraction of problematic regions prevents them from consuming computational resources while maintaining segmentation quality in reliable regions.

Inventive Principle:
Principle #2Taking out (Extraction)

2Manufacturing precision

If non-linear depth adjustment is performed on individual objects, then depth perception quality is improved, but processing time increases

Engineering Contradiction:
Improvedepth perception qualityVSAvoidprocessing time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by quantizing the depth map and identifying depth regions before performing the actual segmentation and depth adjustment. By preparing the depth structure in advance and organizing it into meaningful regions, the subsequent non-linear depth adjustment can be applied efficiently to entire regions rather than individual pixels, reducing processing time while maintaining depth perception quality.

Inventive Principle:
Principle #10Preliminary action

3Reliability

If detailed texture region analysis is performed, then segmentation reliability is improved, but computational complexity increases

Engineering Contradiction:
Improvesegmentation reliabilityVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies local quality by calculating confidence values specifically for texture regions and selectively eliminating only the unreliable ones. Instead of uniformly processing all regions with high computational complexity, the method locally identifies and handles only the problematic regions, maintaining segmentation reliability where needed while reducing overall computational complexity.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP2932466B1Method and apparatus for segmentation of 3D image data
Publication Date: 2019.12.04 HUAWEI TECH CO LTD
  • EP2932466B1 patent drawingFigure 1
  • EP2932466B1 patent drawingFigure 2
  • EP2932466B1 patent drawingFigure 3

AI summary

The present invention provides a method for segmentation of 3D image data of a 3D image, the method comprising: determining (20) local features for each of a plurality of views (101, 102, 103) of the3D image;determining (30) a local feature graph based on the determined local features; and segmenting (40) the 3D image data into a plurality of depth regions based on the determined local feature graph and a depth map (110)of the 3D image.