Multi-scale Feature Hashing for Image Search Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image search methods require direct comparison of images, leading to high comparison workloads and long search times due to limited reference features, resulting in poor search efficiency.
Innovation Solution
An image search method that extracts multi-scale features, determines hash values, and compares them to obtain original images similar to the searched image by grouping them into hash buckets, reducing the number of comparisons needed through the use of a deep hash model with convolutional neural networks and index libraries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct comparison of images is performed in conventional image search methods, then complete image data is available for comparison, but comparison workload increases and search time lengthens
Solution Approach 1:
The patent segments images into multiple feature scales (first scale, second scale, third scale) and extracts features at each scale independently. This segmentation allows the system to compare only relevant feature components rather than entire images, significantly reducing comparison workload while maintaining accuracy. The multi-scale feature extraction divides the complex image comparison task into manageable scale-specific sub-tasks.
Solution Approach 2:
The patent extracts key feature information from images at different scales, separating essential comparative elements from redundant image data. By taking out only the discriminative features at each scale and using them for comparison, the system avoids processing complete images, thereby reducing search time while preserving comparison accuracy through selective feature utilization.
2Reliability
If conventional image comparison methods are used, then all original images in the library are compared, but comparison workload becomes excessively large
Solution Approach 1:
The patent performs preliminary action by pre-processing images to extract multi-scale features and generate hash values before actual search operations. The hash values are used to pre-group images into buckets, so during search, only images within the same hash bucket need comparison. This preliminary organization significantly reduces the number of comparisons required while ensuring no relevant images are missed, thus improving search efficiency without compromising completeness.
Solution Approach 2:
The patent introduces a new dimension by transforming images from pixel-space to feature-space through multi-scale feature extraction. Instead of comparing images directly in their original high-dimensional pixel space, the system projects them into a lower-dimensional feature space organized by hash buckets. This dimensional transformation reduces the search space from all images to only those in relevant hash buckets, dramatically improving productivity while maintaining reliability through the structured feature representation.
3Device complexity
If limited reference features are used in comparison, then comparison process is simpler, but search accuracy deteriorates
Solution Approach 1:
The patent applies local quality by assigning different importance weights to features at different scales. Rather than treating all features uniformly, the system identifies which scales contain more discriminative information for specific comparison tasks and emphasizes those features accordingly. This localized quality adjustment allows the comparison process to focus on the most informative features, maintaining high search accuracy while managing complexity by not processing all features with equal weight.
4Productivity
If interference information is not removed from images, then processing is faster, but search accuracy decreases
Solution Approach 1:
The patent extracts and removes interference information from images before feature extraction by identifying and eliminating background noise, irrelevant objects, and distracting elements. This selective removal of interference preserves the essential target features while reducing noise that would otherwise degrade search accuracy. The processing remains efficient because the removal focuses only on identified interference rather than requiring exhaustive analysis of all image content.
Data Source
AI summary
An image search method and an apparatus may include acquiring an image to be searched; extracting a multi-scale feature of the image to be searched; determining a hash value of the image to be searched according to the multi-scale feature; and obtaining original images similar to the image to be searched by comparing the multi-scale feature of the image to be searched with a multi-scale feature of each original image corresponding to a hash bucket to which the hash value belongs.