Markov Random Field Image Segmentation for Transparent Objects

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing computer vision systems face challenges in accurately segmenting images, particularly in scenes with transparent and reflecting objects, as these objects produce misleading information for RGB and depth sensors, leading to difficulties in material and color transition identification and 3D reconstruction.

Innovation Solution

The method generates a segmented image by combining multiple images from different physical signals, such as RGB, depth, and infrared, using a Markov Random Field (MRF) energy minimization approach, which includes a smoothness term and a data term, to improve segmentation accuracy and robustness, especially in scenes with transparent and reflecting objects.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple physical signals (RGB, depth, infrared) are combined for segmentation, then segmentation accuracy is improved, but device complexity increases

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the segmentation process into distinct functional components: multiple physical signal acquisition modules (RGB camera, depth sensor, infrared sensor) that capture different aspects of the scene, and a Markov Random Field optimization module that integrates these signals. This segmentation allows each sensor to specialize in capturing specific material properties while the MRF model combines them systematically, resolving the contradiction by organizing complexity into manageable segments.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements a multi-functional segmentation system where a single processing framework handles multiple physical signals (RGB for color, depth for geometry, infrared for thermal properties). The Markov Random Field model serves as a universal optimization approach that can process and integrate diverse signal types through a unified energy minimization framework, allowing one system to perform multiple segmentation functions simultaneously.

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

2Measurement precision

If transparent and reflecting objects are segmented using traditional RGB and depth sensors, then processing speed is maintained, but measurement precision deteriorates

Engineering Contradiction:
Improvematerial identification accuracyVSAvoidsensor configuration
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces infrared signal acquisition as an intermediary component that mediates between the challenging transparent/reflecting objects and the segmentation process. Infrared radiation provides complementary information about material thermal properties that are independent of optical transparency or reflectivity, acting as an intermediary signal that reveals material characteristics hidden from RGB and depth sensors alone.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent creates a composite information framework by combining multiple physical signal types (RGB, depth, infrared) into a unified segmentation approach. This composite signal integration allows the system to leverage the strengths of each sensor type: RGB for color information, depth for geometric structure, and infrared for thermal material properties, creating a robust multi-modal representation that overcomes the limitations of individual sensors.

Inventive Principle:
Principle #40Composite materials

Data Source

PatentEP3343504B1Producing a segmented image using markov random field optimization
Publication Date: 2021.12.01 DASSAULT SYSTEMES SA
  • EP3343504B1 patent drawingFigure 1
  • EP3343504B1 patent drawingFigure 2A~2C
  • EP3343504B1 patent drawingFigure 3

AI summary

A computer-implemented method of producing a segmented image of a scene comprises providing a plurality of images of the scene, each image corresponding to a respective acquisition of a physical signal, the plurality of images including at least two images corresponding to different physical signals, and generating the segmented image based on the plurality of images, by determining a distribution of labels that minimizes an energy defined on a Markov Random Field (MRF). This improves the field of computer vision.