Semantic Image Retrieval via Vector Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image retrieval methods fail to identify images that are semantically related to an input keyword unit, even if they are not indexed by that specific keyword unit, due to the limitations of inverted keyword index approaches.
Innovation Solution
The method involves classifying a corpus of digital images into categories using an image classifier, generating image and word vectors based on semantic similarities, and computing vector similarity measurements to retrieve relevant images, allowing for semantic image retrieval even if images are not indexed by the query keyword.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If an inverted keyword index approach is used to retrieve images, then retrieval speed is improved, but retrieval accuracy deteriorates because images semantically related to but not indexed by the query keyword are missed
Solution Approach 1:
The patent introduces word vectors as an intermediary between the query keyword and image retrieval. Instead of directly matching keywords in an inverted index, the system converts the query keyword into a word vector and retrieves images based on vector similarity. This intermediary enables semantic matching beyond exact keyword matches, resolving the contradiction between fast indexed retrieval and accurate semantic retrieval.
2Ease of manufacture
If exact keyword matching is used for image retrieval, then implementation simplicity is improved, but retrieval completeness deteriorates because semantically related images without exact keyword matches are not retrieved
Solution Approach 1:
The patent transforms the retrieval parameter from discrete exact keyword matching to continuous vector similarity measurement. By representing keywords and images as vectors in a semantic space, the system can measure similarity through continuous distance metrics rather than binary exact matches. This parameter change enables retrieval of semantically related images while maintaining computational feasibility through vector operations.
Data Source
AI summary
Computer-implemented techniques for sematic image retrieval. According to one technique, digital images are classified into N number of categories based on their visual content. The classification provides a set of N-dimensional image vectors for the digital images. Each image vector contains up to N number of probability values for up to N number of corresponding categories. An N-dimensional image match vector is generated that projects an input keyword query into the vector space of the set of image vectors by computing the vector similarities between a word vector for the input query and a word vector for each of the N number of categories. Vector similarities between the image match vectors and the set of image vectors can be computed to determine images semantically relevant to the input query.