Content Adaptive Hash Lookups for Near-Duplicate Image Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting near-duplicate images in large databases are inefficient due to long search times, especially when images are scaled or modified through re-encoding, printing, scanning, resizing, or other transformations, leading to undesirably high search times as the database size increases.
Innovation Solution
A scalable and high-performance near-duplicate image search method using short hashes that performs content adaptive hash lookups by adjusting the reliability of each hash bit, allowing detection of cropped, resized, and re-encoded images by incorporating local features and scale invariance, and using a key-value table with reliability-based lookups to compensate for image modifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional image matching algorithms are used to detect near-duplicate images in a large database, then the detection can identify duplicate images, but the search time grows significantly as the database size increases
Solution Approach 1:
The patent segments the image matching process into multiple stages: first extracting interest points and computing short hashes to create compact signatures, then using these signatures for rapid database searching, and finally performing detailed feature comparison only on candidate matches. This segmentation allows the system to handle large databases efficiently by filtering out non-matching images early in the process.
Solution Approach 2:
The patent changes the parameter representation from full image data or detailed feature vectors to compact hash signatures (e.g., 32-bit or 64-bit integers). This parameter transformation enables rapid comparison and indexing operations while preserving the essential information needed to identify near-duplicate images, thereby reducing search time in large databases.
2Productivity
If exact hash matching is used for image search, then the lookup speed is fast, but the method fails to detect near-duplicate images that have been modified by re-encoding, printing, scanning, or resizing
Solution Approach 1:
The patent introduces dynamic adaptivity into the hash matching process by computing reliability scores for each hash bit based on the stability of the corresponding feature. The search algorithm dynamically adjusts the matching threshold and tolerance based on these reliability scores, allowing it to accommodate modifications such as re-encoding, printing, scanning, and resizing while maintaining efficient lookup performance.
Solution Approach 2:
The patent implements a feedback mechanism where the reliability of each hash bit is computed based on the stability of the underlying feature across different images. This reliability information feeds back into the search process, allowing the system to adjust its matching criteria dynamically. Features that are stable across transformations generate reliable hash bits that are weighted more heavily, while unstable features generate less reliable bits that are tolerated more freely.
3Measurement precision
If more detailed feature vectors are used to improve detection accuracy, then the ability to detect modified images improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent extracts only the most essential and stable features from images to create compact signatures. Instead of using full feature vectors, the system identifies interest points (such as corners or distinctive structures) and extracts a small set of invariant properties from each point. These extracted features are then quantized into hash bits, achieving high discriminative power with minimal computational complexity.
Solution Approach 2:
The patent uses short, compact hash signatures (e.g., 32-bit or 64-bit integers) as disposable representations of complex image features. These short hashes serve as efficient proxies for detailed feature vectors, enabling rapid comparison and indexing operations. The full feature extraction and detailed comparison are performed only when necessary, reducing overall computational complexity while maintaining detection accuracy.
Data Source
AI summary
A scalable and high performance near-duplicate image search method utilizing short hashes improves performance over existing methods. By leveraging the shortness of the hashes, the search algorithm analyzes the reliability of each bit of a hash and performs content adaptive hash lookups by adaptively adjusting the “range” of each hash bit based on reliability. Matched features are post-processed to determine the final match results. The method can detect cropped, resized, print-scanned and re-encoded images and pieces from images among thousands of images.