Scale Space Object Boundary Identification in Image Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current image processing techniques, such as rotoscoping and alpha-channel algorithms, face challenges in accurately identifying object boundaries, especially in complex backgrounds and noisy environments, often requiring skilled users and failing to produce hard segmentations effectively.

Innovation Solution

The method involves generating scale space images from an input image using filters like Gaussian kernels or wavelet filters, which create multi-resolution representations to determine potential values for each pixel, forming a potential map that indicates the likelihood of a pixel being within or outside an object boundary, and using this map to identify and refine object boundaries.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional rotoscoping techniques are used to identify object boundaries, then object boundaries can be extracted, but the process becomes labor-intensive and requires skilled users

Engineering Contradiction:
Improveobject boundary identification accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system performs automatic object boundary identification using scale space representations and potential maps, eliminating the need for skilled user intervention. The algorithm independently processes images to generate hard segmentations and identify object boundaries without manual rotoscoping, making the system self-sufficient in performing the segmentation task.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces manual mechanical rotoscoping operations with an automated computational system. Instead of skilled users manually tracing boundaries frame-by-frame, the system uses scale space image processing, potential map generation, and automatic segmentation algorithms to identify object boundaries efficiently and accurately.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If alpha-channel algorithms are used to extract soft boundaries, then color information can be analyzed, but hard segmentations are not produced effectively

Engineering Contradiction:
Improveboundary extraction accuracyVSAvoidhard segmentation quality
Core Design Contradiction:
Measurement precisionVSManufacturing precision

Solution Approach 1:

The system transforms the soft boundary information from alpha-channel algorithms into hard segmentation by changing the parameter representation. Instead of working directly with soft alpha values, the patent generates potential maps and uses thresholding or optimization to convert these continuous values into discrete hard segmentations, effectively transforming the nature of the boundary representation.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces potential maps as an intermediary between alpha-channel analysis and hard segmentation. The potential maps serve as a bridge that translates soft boundary information into a form that can be effectively converted into hard segmentations, mediating between the continuous color analysis and discrete object separation.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If contour-based methods with stroke matching are used, then correspondence between contours can be established, but skilled user input is still required for accurate delineation

Engineering Contradiction:
Improvecontour correspondence accuracyVSAvoiduser skill requirement
Core Design Contradiction:
ReliabilityVSEase of operation

Solution Approach 1:

The system automatically establishes contour correspondence between frames without requiring skilled user input for stroke matching. The scale space representation and potential map generation processes independently identify and track object boundaries across frames, making the contour correspondence task self-service rather than user-dependent.

Inventive Principle:
Principle #25Self-service

4Adaptability or versatility

If random walks algorithm is used for image segmentation, then probability-based segmentation can be achieved, but accurate boundary identification fails in noisy environments

Engineering Contradiction:
Improvesegmentation flexibilityVSAvoidboundary identification accuracy in noise
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent converts the harmful effect of noise into a benefit by using scale space representations that inherently filter noise through multi-resolution analysis. The potential map generation process leverages the noise characteristics to enhance boundary contrast, transforming what would be interference into useful signal for more accurate boundary identification.

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

Data Source

PatentEP2374107B1Devices and methods for processing images using scale space
Publication Date: 2020.07.08 IMAX CORP
  • EP2374107B1 patent drawingFigure 1
  • EP2374107B1 patent drawingFigure 2
  • EP2374107B1 patent drawingFigure 3

AI summary

Certain embodiments relate to processing images by creating scale space images from an image and using them to identify boundaries of objects in the image. The scale space images can have varying levels of detail. They are used to determine a potential map, which represents a likelihood for pixels to be within or outside a boundary of an object. A label estimating an object boundary can be generated and used to identify pixels that potentially may be within the boundary. An image with object boundaries identified can be further processed before exhibition. For example, the images may be two-dimensional images of a motion picture. Object boundaries can be identified and the two- dimensional (2D) images can be processed using the identified object boundaries and converted to three-dimensional (3D) images for exhibition.