Image Descriptor Generation Using Texture Intensity Weighting

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing methods fail to accurately identify images that are similar but not identical, as they focus on exact matches, neglecting occlusion and objects within images, and require significant data processing, making it difficult to retrieve images that are perceptually similar to an input image.

Innovation Solution

A method is developed to generate a descriptor for at least part of an image by processing image data to identify texture characteristics, weighting them with intensity data, and applying filters to capture texture and intensity information across multiple scales and symmetries, allowing for the differentiation of similar image regions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If known image processing methods are used to identify similar images, then exact matches can be identified, but images that are perceptually similar but not identical cannot be accurately identified

Engineering Contradiction:
Improveimage matching accuracyVSAvoidability to identify perceptually similar images
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The image is divided into multiple patches, and each patch is processed independently to extract texture features. This segmentation allows the method to capture local texture characteristics that may differ between identical and perceptually similar images, thereby improving the ability to identify perceptually similar images while maintaining exact match accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies multiple texture filters (e.g., Gabor filters, Laplacian filters) with different parameters to extract texture features at multiple scales and orientations. By changing filter parameters and combining results, the method achieves both precise matching for identical images and adaptability for perceptually similar images.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If multiple texture filters are applied to extract texture features, then texture information is captured accurately, but data processing requirements increase significantly

Engineering Contradiction:
Improvetexture characterization accuracyVSAvoiddata processing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The image is divided into multiple patches, and each patch is processed independently to extract texture features. This segmentation allows the method to capture local texture characteristics that may differ between identical and perceptually similar images, thereby improving the ability to identify perceptually similar images while maintaining exact match accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies multiple texture filters (e.g., Gabor filters, Laplacian filters) with different parameters to extract texture features at multiple scales and orientations. By changing filter parameters and combining results, the method achieves both precise matching for identical images and adaptability for perceptually similar images.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If complete image descriptors are generated, then overall image similarity can be assessed, but occlusion and objects within images cannot be accounted for

Engineering Contradiction:
Improveoverall image similarity assessmentVSAvoidocclusion and object detection
Core Design Contradiction:
Measurement precisionVSDifficulty of detecting and measuring

Solution Approach 1:

The image is divided into multiple patches, and each patch is processed independently to extract texture features. This segmentation allows the method to capture local texture characteristics that may differ between identical and perceptually similar images, thereby improving the ability to identify perceptually similar images while maintaining exact match accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Different texture filters are applied to different patches based on local characteristics. The method extracts texture features from each patch independently and combines them, allowing occluded regions or specific objects to be detected through their unique texture patterns while maintaining overall image similarity assessment.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3044733B1Image processing
Publication Date: 2024.10.30 ZEBRA TECHNOLOGIES CORP
  • EP3044733B1 patent drawingFigure 1
  • EP3044733B1 patent drawingFigure 2
  • EP3044733B1 patent drawingFigure 3

AI summary

A method of generating a descriptor of at least part of an image includes receiving image data representing the at least part of the image. The image data is processed to identify at least one texture characteristic of the at least part of the image, thereby generating texture data indicative of a texture of the at least part of the image. The texture data is processed with the image data, thereby generating weighted texture data. A descriptor of the at least part of the image is generated using the weighted texture data.