Image Feature Search Engine Visual Similarity Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image search engines lack the ability to effectively utilize complex visual attributes and cannot search unindexed image repositories, limiting their ability to provide accurate and relevant search results based on visual similarity.
Innovation Solution
The implementation of an Image Feature Search Engine (IFSE) that selects images for comparison, specifies visual criteria based on similarity, searches images using these criteria, and displays subsets that meet a visual similarity threshold, incorporating features like overall composition, color, and luminosity, through a system comprising a Feature Detection Module, Criteria Weighting Module, and Image Selection and Ordering Module.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If image search is based on simple attributes like color or pixel dimensions, then search speed is improved, but search accuracy and relevance are worsened
Solution Approach 1:
The patent segments the image search process into multiple stages: initial filtering based on simple attributes (color, dimensions) and subsequent refinement based on complex visual attributes (composition, luminosity, texture). This allows the system to maintain fast initial processing while achieving high accuracy through progressive filtering and scoring.
Solution Approach 2:
The system performs preliminary indexing of both simple and complex visual attributes during image ingestion. This preliminary action enables the search engine to quickly retrieve candidate images using simple attributes while having complex attributes pre-computed and ready for rapid comparison during the search phase, eliminating the need for time-consuming real-time analysis.
2Measurement precision
If image search uses complex visual attributes like composition and luminosity, then search accuracy is improved, but computational complexity is worsened
Solution Approach 1:
The patent transforms complex visual attributes into standardized numerical parameters that can be efficiently computed and compared. For example, composition is represented as spatial relationships between detected objects, luminosity as histograms of brightness values, and texture as frequency domain features. These parameter transformations enable complex comparisons to be performed using efficient mathematical operations rather than computationally intensive image processing.
Solution Approach 2:
The system creates simplified representations (copies) of complex visual attributes during the indexing phase. Instead of storing and processing actual image data during search, it uses pre-computed feature vectors and descriptors that capture the essential characteristics of composition, luminosity, and texture. This copying approach allows rapid comparison without repeatedly analyzing the original complex images.
3Measurement precision
If images are indexed with specific attributes, then search relevance is improved, but adaptability to unindexed repositories is worsened
Solution Approach 1:
The patent implements a universal image analysis framework that can extract and evaluate multiple types of visual attributes (color, composition, luminosity, texture, objects) from any image regardless of its source or existing index. This multi-functional capability allows the same search engine to effectively query both indexed repositories with pre-computed attributes and unindexed repositories by performing on-the-fly attribute extraction, providing consistent search relevance across diverse data sources.
Data Source
AI summary
Provided are techniques for selecting an image for comparison with a plurality of images; specifying one or more visual criterion for searching the plurality of images, wherein the visual criteria are based upon a visual similarity between the image and the plurality of images; searching the plurality of images based upon the criteria to identify a subset of the plurality of images that meet a threshold with respect to the visual similarity; and displaying in a graphical user interface the subset of images.