Picture Search Using Tiered OCR for Fast, Accurate Text Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing picture search methods based on optical character recognition (OCR) are inefficient due to high time costs for complete OCR processing, especially when dealing with large numbers of pictures, and fail to accurately identify detailed text information, while cloud recognition faces risks of failure and privacy concerns.
Innovation Solution
A picture search method that combines low-dimensional and high-dimensional OCR processing, where low-dimensional OCR is performed based on an OCR threshold to quickly identify easily recognizable text, and high-dimensional OCR is used for detailed recognition, with results stored for subsequent matching, optimizing search efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complete OCR processing is performed on all pictures, then recognition accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent segments the OCR processing into two distinct levels: low-dimensional OCR for quick initial recognition and high-dimensional OCR for detailed accurate recognition. This segmentation allows the system to first quickly identify obviously matching pictures using low-dimensional processing, then apply high-dimensional processing only to borderline cases, thereby resolving the contradiction between speed and accuracy.
Solution Approach 2:
The patent applies partial action by performing high-dimensional OCR processing only on pictures that fall within a preset accuracy range after low-dimensional processing, rather than processing all pictures with high-dimensional OCR. This selective approach maintains high recognition accuracy for matching pictures while significantly reducing overall processing time.
2Productivity
If low-dimensional OCR is used for all pictures, then processing speed is improved, but recognition accuracy decreases
Solution Approach 1:
The patent uses low-dimensional OCR processing for all pictures as a first pass to maintain high processing speed, then applies high-dimensional OCR processing selectively to pictures that fall within a preset accuracy range. This ensures that processing speed is not compromised while still achieving high recognition accuracy for final results.
Solution Approach 2:
The system implements feedback by using the results of low-dimensional OCR processing to determine which pictures require further high-dimensional processing. Pictures with recognition results within the preset accuracy range are selected for high-dimensional processing, ensuring that accuracy is improved where needed while maintaining overall efficiency.
3Measurement precision
If high-dimensional OCR is applied to all pictures, then recognition accuracy is improved, but computational resources are excessively consumed
Solution Approach 1:
The patent segments the picture set into two groups: pictures processed only by low-dimensional OCR and pictures requiring high-dimensional OCR. This segmentation is based on whether the low-dimensional OCR result falls within the preset accuracy range, allowing computational resources to be concentrated only where necessary for high accuracy.
Solution Approach 2:
The patent applies high-dimensional OCR processing only partially to pictures that require it, rather than applying it universally. This selective application significantly reduces computational resource consumption while maintaining high recognition accuracy for the final search results.
Data Source
AI summary
A picture search method and apparatus, an electronic device, a computer-readable storage medium, and a computer program product, relating to the field of artificial intelligence that can obtain an OCR result of pictures in a preset picture library in response to a picture search request; traverse pictures which are not subjected to low-dimensional OCR processing and high-dimensional OCR processing in the preset picture library, and perform the low-dimensional OCR processing based on an OCR threshold on each of the traversed pictures to obtain a low-dimensional OCR result of each corresponding picture; determining a target picture matching a key character string in the preset picture library according to at least one of the low-dimensional OCR result and the high-dimensional OCR result of each picture; and determining the target picture as a search result of the picture search request, and displaying the search result.