Image Segmentation via Adaptive Confidence Thresholding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing segmentation methods face challenges in achieving high precision and processing efficiency, particularly in identifying individual instances of object classes in images.

Innovation Solution

The method focuses on fine-tuning confidence score thresholds within coherent regions, which are identified by analyzing motion vectors, to generate precise segmentation masks. This approach reduces processing resources by minimizing threshold adjustments in non-object areas.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If confidence score thresholds are fine-tuned across the entire image, then segmentation precision is improved, but processing time and computational resources increase

Engineering Contradiction:
Improvesegmentation precisionVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The image is divided into multiple regions of interest based on motion vector analysis. Threshold fine-tuning is applied selectively to each region rather than uniformly across the entire image, reducing computational overhead while maintaining segmentation precision in areas where objects are likely to be present

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Different threshold adjustment strategies are applied to different regions of the image based on local characteristics. Regions with high motion coherence receive more aggressive threshold fine-tuning, while static regions use simpler thresholding, optimizing the balance between precision and processing efficiency

Inventive Principle:
Principle #3Local quality

2Measurement precision

If threshold adjustments are applied to all image areas, then segmentation accuracy is improved, but processing efficiency deteriorates

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

Motion vector analysis and region of interest identification are performed as preliminary steps before threshold fine-tuning. This pre-processing identifies areas where objects are likely present, allowing subsequent threshold adjustments to be focused only on these regions, thereby improving processing efficiency without sacrificing segmentation accuracy

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Instead of applying threshold fine-tuning uniformly across the entire image, the method applies partial action by focusing computational resources only on identified regions of interest. This selective approach maintains segmentation accuracy for potential objects while significantly reducing overall processing time

Inventive Principle:
Principle #16Partial or excessive action

3Area of stationary object

If processing resources are allocated uniformly across the image, then coverage is improved, but resource utilization efficiency worsens

Engineering Contradiction:
Improveimage coverageVSAvoidresource utilization efficiency
Core Design Contradiction:
Area of stationary objectVSLoss of energy

Solution Approach 1:

Processing resources are allocated non-uniformly based on local image characteristics. Regions identified as potential object locations through motion vector analysis receive higher resource allocation and more aggressive threshold fine-tuning, while static background regions receive minimal processing, optimizing resource utilization efficiency

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4156098B1A segmentation method
Publication Date: 2025.05.14 AXIS
  • EP4156098B1 patent drawingFigure 1~2
  • EP4156098B1 patent drawingFigure 3a~3c
  • EP4156098B1 patent drawingFigure 4~5

AI summary

A method of generating a segmentation outcome which indicates individual instances of one or more object classes for an image in a sequence of images is disclosed. The method comprises: determining (501) a coherent region of the image; processing (502) the image to determine a tensor representing pixel-specific confidence scores; generating (503) a series of temporary segmentation masks for the coherent region, wherein each temporary segmentation mask is generated by interpreting the tensor with respect to a single object class using a different temporary confidence score threshold; evaluating (504) the series of temporary segmentation masks to determine if an object mask condition is met; depending on the outcome of the evaluation, setting (505) the temporary confidence score threshold as a final confidence score threshold for the pixels of the temporary segmentation mask, or setting (505) a default confidence score threshold as a final confidence score threshold for the coherent region; and generating (506) a final segmentation outcome for the image.