Image Descriptor Generation Using Texture Intensity Weighting
Find Innovative SolutionsGenerate 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
Engineering 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
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.
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.
2Measurement precision
If multiple texture filters are applied to extract texture features, then texture information is captured accurately, but data processing requirements increase significantly
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.
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.
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
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.
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.
Data Source
Figure 1
Figure 2
Figure 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.