Re-ranking Object Recognition Results by Text Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for providing users with information via computing devices are inefficient, as they require manual input and struggle to accurately identify relevant product information from images, often leading to tedious tasks and incorrect search results.
Innovation Solution
A system that processes images using optical character recognition (OCR) to identify and filter relevant text, generating a search query for electronic marketplaces, and re-ranking search results based on semantic analysis and model number detection to provide accurate product information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional OCR is used to extract text from images for product search, then text extraction is achieved, but irrelevant text and false positives are included in search results
Solution Approach 1:
The patent changes the parameters of text evaluation by assigning different weights to different text elements based on their relevance to product identification. Model numbers receive higher weights than generic descriptors, and text proximity to product images influences weighting. This differential weighting resolves the contradiction by filtering irrelevant text while preserving meaningful product identifiers in search results.
Solution Approach 2:
The patent creates a weighted copy of the extracted text where each word or phrase is replicated with a weight value reflecting its relevance to product identification. This weighted text copy is then used for search queries, allowing the system to emphasize critical identifiers while de-emphasizing irrelevant information, thereby improving both accuracy and reliability.
2Productivity
If all extracted text is used for search queries, then comprehensive search coverage is achieved, but search efficiency and accuracy decrease due to irrelevant information
Solution Approach 1:
The patent applies parameter changes by transforming raw extracted text into weighted text where each element carries a relevance score. This transformation enables the search system to process fewer, more relevant terms efficiently while maintaining high identification accuracy. The weighting parameter acts as a filter that improves both productivity and precision simultaneously.
Solution Approach 2:
The patent extracts only the most relevant text elements for search queries by applying weight thresholds and relevance criteria. Instead of using all extracted text, the system selectively extracts high-value identifiers such as model numbers and product-specific terms, discarding irrelevant information. This selective extraction improves search efficiency without sacrificing identification accuracy.
3Ease of operation
If manual text input methods are used for product search, then user control is maintained, but user effort and time consumption increase significantly
Solution Approach 1:
The patent enables the system to perform product identification automatically by capturing images and processing the text content without requiring manual user input. The system self-services the entire workflow from image capture through text extraction, weighting, and search execution, dramatically reducing both user effort and time consumption while maintaining ease of operation.
Solution Approach 2:
The patent performs preliminary text extraction, filtering, and weighting actions automatically before the user needs to search for products. By pre-processing the image content and preparing weighted search terms in advance, the system eliminates the need for manual text input and accelerates the product finding process, resolving the contradiction between ease of operation and time loss.
Data Source
AI summary
A user can capture an image of a text object of interest and have that image submitted for processing. The image can be pre-processed to improve quality and then submitted to an optical character recognition process to identify the words, characters, or strings in the image. At least some of these results can be submitted as a query to a search engine to obtain potential matches. In order to improve the accuracy of the results, information such as the titles for the results can be compared against each recognized word, character, or string from the image, including the ordering of those elements. An updated relevancy score can then be generated based on the full, ordered set. The recognized text is also analyzed to attempt to recognize model numbers or other identifiers that can be weighted more heavily as being indicative of accurate matches. Matches are selected from the re-ranked results.