Near Duplicate Image Detection Using Visual Word Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image search systems face challenges in efficiently identifying and removing near-duplicate images from search results, which can clutter the output and affect query response times.
Innovation Solution
The system generates feature vectors for each image tile, using a vocabulary of visual words to classify images as near-duplicates by matching visual words and color descriptors, allowing for rapid comparison and online determination of near-duplicates without significant computational or storage costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image search systems compare all images to identify near-duplicates, then identification accuracy improves, but query response time increases
Solution Approach 1:
The patent segments images into multiple tiles and extracts visual words from each tile independently. This segmentation allows the system to compare only relevant portions of images rather than entire images, improving identification accuracy while reducing computational time. The feature descriptor combines visual words with their spatial locations to enable efficient near-duplicate detection.
Solution Approach 2:
The patent performs preliminary computation of feature descriptors and stores them in an index before actual search queries. This preliminary action pre-processes images into compact representations that can be quickly compared during search operations, significantly reducing query response time while maintaining identification accuracy.
2Measurement precision
If detailed feature comparison is performed to accurately identify near-duplicates, then identification accuracy improves, but computational cost increases
Solution Approach 1:
The patent extracts key visual features from images and represents them as a compact set of visual words in a vocabulary. This extraction process converts complex image data into a simplified representation that retains essential information for near-duplicate identification while dramatically reducing computational requirements for comparison.
Solution Approach 2:
The patent changes the parameter representation from continuous pixel values to discrete visual word indices with spatial coordinates. This parameter transformation enables efficient comparison operations and reduces computational cost while maintaining the ability to accurately identify near-duplicates through the feature descriptor structure.
3Reliability
If comprehensive image comparison is performed across different collections, then near-duplicate detection completeness improves, but storage requirements increase
Solution Approach 1:
The patent creates compact copies of image features in the form of feature descriptors that store only essential information (visual words and spatial locations) rather than the full image data. These compact representations can be stored efficiently and used for cross-collection near-duplicate detection without requiring significant storage resources.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for determining image search results. One of the methods includes generating a plurality of feature vectors for each image in a collection of images, wherein each feature vector is associated with an image tile of an image, wherein each feature vector corresponds to one of a plurality of predetermined visual words. All images in the collection of images that share at least a threshold number of matching visual words associated with matching image tiles are classified as near-duplicate images.