Visual Search Keypoint Selection Using Color Factor Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current visual search methods face challenges in effectively combining grayscale and color information for local descriptor selection and global descriptor generation, leading to suboptimal performance in image retrieval.
Innovation Solution
The method involves extracting gray-keypoints and color-keypoints from query images, computing a color factor to determine the amount of color gradients, and combining these keypoints to enhance local descriptor selection and global descriptor generation, using techniques such as convex hull distinction, visual meaning scoring, and quantization error-based selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If only grayscale keypoints are extracted for visual search, then the processing complexity is reduced, but the search accuracy deteriorates due to lack of color information
Solution Approach 1:
The patent applies local quality by selectively extracting color-keypoints only from regions where color information is meaningful and useful for the search task. Instead of processing the entire image with color data, the system identifies specific local regions (through grayscale keypoint detection first) and then extracts color information only from those regions, thereby reducing overall processing complexity while maintaining search accuracy.
Solution Approach 2:
The patent segments the keypoint extraction process into two distinct stages: first extracting grayscale keypoints to identify potential regions of interest, then selectively extracting color-keypoints only from those identified regions. This segmentation allows the system to handle color information only where necessary, reducing computational complexity while preserving accuracy.
2Quantity of substance
If color-keypoints are extracted from all color channels, then the color information completeness is improved, but the computational time increases
Solution Approach 1:
The patent applies local quality by extracting color-keypoints only from specific local regions where grayscale keypoints were detected, rather than processing all color channels across the entire image. This selective extraction maintains color information completeness in relevant regions while significantly reducing computational time.
Solution Approach 2:
The patent performs preliminary grayscale keypoint extraction before color keypoint extraction. This preliminary action identifies the regions of interest in advance, allowing the subsequent color keypoint extraction to be performed only in those specific regions, thereby reducing computational time while maintaining information completeness.
3Device complexity
If local descriptors are selected based on scale threshold alone, then the selection process is simplified, but the descriptor relevance deteriorates
Solution Approach 1:
The patent merges multiple selection criteria into a unified local descriptor selection process. Instead of using scale threshold alone, the system combines scale information with color factor values and visual meaning scores to comprehensively evaluate and select local descriptors, thereby improving descriptor relevance while maintaining reasonable process complexity.
Solution Approach 2:
The patent uses a composite evaluation approach where local descriptors are assessed based on multiple factors (scale, color factor, visual meaning score) simultaneously. This composite material approach to descriptor selection ensures that descriptors with high relevance are identified, combining the strengths of different evaluation metrics.
4Measurement precision
If convex hull is calculated for all local descriptors, then the foreground distinction is improved, but the computational load increases
Solution Approach 1:
The patent applies local quality by calculating the convex hull only for local descriptors that are candidates for foreground extraction, rather than processing all descriptors in the image. This selective approach improves foreground distinction accuracy in relevant regions while reducing overall computational load.
Solution Approach 2:
The patent uses partial action by computing the convex hull only for a subset of local descriptors that meet certain criteria (such as scale threshold or visual meaning score), rather than processing all descriptors. This partial computation achieves sufficient foreground distinction without the excessive computational cost of full-image processing.
Data Source
AI summary
A method comprises computing a color factor value indicating an amount of color gradients in at least one color channel from the query image. The method comprises combining the color-keypoints with the gray-keypoints when the color factor value is greater than a threshold. A method for performing a visual search comprises extracting a plurality of local descriptors from a query image and then selecting a subset of them based on various criteria's such as visual meaning score. A method comprises aggregating each mean vector for each visual codeword from distances between each visual codeword and local descriptors. The method comprises aggregating variance vector for each visual codeword from the distance between each visual codeword, and local descriptors. The method comprises transmitting aggregated mean vector information and aggregated variance vector information to a search server for efficient image retrieval.