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 on a mannequin, as they assume physical separation between foreground and background, leading to issues with occlusion and color spill, and lack methods for accurate body shape generation and surface geometry for outfit visualization.
Innovation Solution
A method and system that measure the attenuation of garment fabric light between the mannequin and camera to compute alpha mattes, allowing for the generation of alpha mattes with a background that includes objects in the foreground, using different spectral power distributions for image acquisition, and a system for generating a body model by defining control points and measurements to align and warp garment images onto a body model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing image processing techniques are used to generate alpha mattes, then the process is simple, but the accuracy is poor when foreground objects overlap or are close to the background
Solution Approach 1:
The patent segments the image into multiple layers (foreground objects at different depths and background) and processes each layer separately using depth information. This allows accurate alpha matte generation for overlapping objects by treating each object independently rather than as a single composite foreground.
Solution Approach 2:
The patent introduces depth information (z-dimension) to the traditional 2D image processing problem. By incorporating depth maps or depth estimation, the system can distinguish between overlapping foreground objects and background elements that appear similar in 2D, thereby improving alpha matte accuracy without significantly increasing processing complexity.
2Adaptability or versatility
If foreground objects are positioned close to the background, then scene realism is improved, but discrimination between foreground and background becomes difficult
Solution Approach 1:
By adding depth information as an additional dimension, the patent enables reliable foreground-background discrimination even when objects are positioned close together in the 2D image plane. The depth dimension provides the necessary separation information that is not visible in standard 2D images.
Solution Approach 2:
The patent changes the parameter space by incorporating depth values alongside color and intensity information. This allows the system to distinguish between foreground and background objects based on their depth parameters, maintaining scene realism while solving the discrimination problem.
3Measurement precision
If manual user intervention is used for alpha matte computation, then accuracy is improved, but processing time and cost increase
Solution Approach 1:
The patent implements automated alpha matte computation that performs the segmentation and depth analysis without requiring manual user intervention. The system uses algorithms to automatically identify foreground objects, estimate depth, and generate alpha mattes, thereby maintaining accuracy while eliminating the time cost of manual processing.
Solution Approach 2:
The patent replaces manual mechanical intervention with automated computational algorithms. By using computer vision techniques and depth estimation algorithms, the system achieves accurate alpha matte generation automatically, substituting human effort with computational processes that are both accurate and efficient.
4Ease of manufacture
If conventional lighting methods are used, then setup is simple, but color spill and lighting interference occur when objects are close together
Solution Approach 1:
The patent applies different lighting characteristics to different spatial regions and depth layers. By controlling lighting locally for foreground objects versus background, and using depth information to separate their contributions, the system eliminates color spill while maintaining simple overall lighting setup. Each object receives appropriate lighting treatment based on its depth position.
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 reduced user intervention, accurate segmentation of garments from mannequins, and precise generation of body models for realistic outfit visualization, addressing the limitations of existing techniques in occlusion and color handling.
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
Implementation Method 3
measure the attenuation of the garment fabric of light between points on the surface of the mannequin or backdrop and the camera
Data Source
Figure 1a
Figure 1b
Figure 2
AI summary
Images of foreground objects in a scene are generated by causing electromagnetic radiation to be emitted having a first spectral power distribution from a surface of a first foreground object, which is adjacent or at least partially obscured by a second foreground object. A first image of both of the first and second foreground objects is acquired whilst the first foreground object emits electromagnetic radiation with the first spectral power distribution. A second image of the first and second foreground objects is acquired whilst the first foreground object is not emitting electromagnetic radiation or is emitting electromagnetic radiation with a second spectral power distribution which is different to the first spectral power distribution. An alpha matte of the first and second foreground objects is generated based on a comparison of the first image and second image.