System and method for image processing and generating a body model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques fail to accurately generate alpha mattes for garments worn on a mannequin, as they struggle with partial opacity and misalignment issues due to the mannequin being part of the foreground, leading to challenges in separating garments from the background and generating accurate body models.
Innovation Solution
A method and system that measure the attenuation of light between the mannequin and camera, allowing for the generation of alpha mattes with a background that includes objects in the foreground, using different spectral power distributions for image acquisition to distinguish between the mannequin and garments, and a process for generating accurate body models by defining control points and measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing image processing techniques are used to generate alpha mattes for garments on a mannequin, then the process is simple and fast, but the accuracy of separating garments from the mannequin and background is poor due to partial opacity and misalignment issues
Solution Approach 1:
The image processing is divided into multiple specialized stages: foreground segmentation to separate garments from background, mannequin segmentation to separate mannequin from garments, and alpha matte generation. Each stage addresses specific challenges with targeted processing, improving overall accuracy while managing complexity through modular organization.
Solution Approach 2:
A synthetic image is generated as an intermediary representation that combines information from multiple input images. This synthetic image serves as a mediator that facilitates the separation of garments, mannequin, and background by encoding their respective contributions, enabling more accurate alpha matte generation without directly solving the complex segmentation problem in a single step.
2Measurement precision
If the mannequin is treated as background, then the processing is simpler, but the accuracy deteriorates because the mannequin is physically part of the foreground and causes occlusion and lighting interactions
Solution Approach 1:
The scene is segmented into three distinct foreground components: garments, mannequin, and background. By treating the mannequin as a separate foreground object rather than background, the system accurately captures its occlusion effects and lighting interactions while maintaining processing simplicity through automated segmentation algorithms.
Solution Approach 2:
The system changes the classification parameter of the mannequin from background to foreground object. This parameter change enables accurate modeling of the mannequin's physical properties including occlusion, transparency, and lighting interactions, while the automated processing maintains ease of operation.
3Measurement precision
If multiple images are acquired with different spectral power distributions, then the ability to distinguish between mannequin and garments is improved, but the acquisition time and complexity increase
Solution Approach 1:
The system uses periodic illumination with different spectral power distributions to illuminate the scene, capturing multiple images at different time points. This periodic action enables spectral differentiation between garments and mannequin while managing acquisition time through efficient sequencing of illumination cycles.
Solution Approach 2:
The spectral power distribution parameter of the illumination is changed across multiple acquisitions, enabling the system to distinguish between materials with different spectral signatures. This parameter variation provides the necessary information for accurate segmentation while the automated processing minimizes the time loss.
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
Enables automatic computation of alpha mattes with improved accuracy for garments on a mannequin, addressing partial opacity and misalignment issues, and allows for precise generation of body models, enhancing realism in image compositing and outfit visualization.
Implementation Method 1
emitting or retroreflecting electromagnetic radiation having a first spectral power distribution from a surface of a first foreground object
Implementation Method 2
emitting or retroreflecting electromagnetic radiation having a first spectral power distribution from a surface of a first foreground object
Implementation Method 3
acquiring a first image of both the first and second foreground objects whilst the first foreground object retroreflects or emits electromagnetic radiation
Data Source
Figure 1a
Figure 1b
Figure 2
AI summary
Manipulated image sprites are generated by defining at least one sprite control point on a garment image sprite, defining, in the body model image, at least one output control point corresponding to each sprite control point, generating a mapping of each sprite control point to each output control point and manipulating pixels of the image sprite based on the mapping so that pixels in the image sprite align with pixels in the body image.