Partial Document Recognition for Complete Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine recognition systems such as OCR and ASR are prone to errors due to imperfect scanning, processing, and flawed algorithms, making it difficult to accurately recognize and retrieve complete documents, especially when portions are damaged or of poor quality.
Innovation Solution
A system and method that uses machine recognition outputs as input for search engines to locate complete documents by scanning or recognizing portions of text or speech, allowing refinement of recognition results and enabling retrieval of entire documents from databases or networks, even when only partial information is available.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If machine recognition systems (OCR/ASR) are used to recognize text or speech from damaged or poor quality sources, then document retrieval is enabled, but recognition accuracy deteriorates due to errors from imperfect scanning, processing, and flawed algorithms
Solution Approach 1:
The patent introduces an intermediary search system that bridges the gap between imperfect machine recognition outputs and complete document retrieval. The search system acts as a mediator that takes the erroneous recognized portions as input queries, searches the database for matching documents, and returns complete copies. This intermediary layer enables document retrieval functionality while circumventing the accuracy limitations of the underlying machine recognition systems.
Solution Approach 2:
The system implements feedback by using the search results (complete document copies) to refine and correct the original machine recognition outputs. The recognized portions are compared against the retrieved complete documents, allowing error correction and improvement of recognition accuracy through iterative feedback from the search and retrieval process.
2Productivity
If only partial portions of documents are scanned for recognition, then processing time and resource consumption are reduced, but completeness of retrieved information deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the document processing into two distinct stages: (1) scanning and recognizing only partial portions of documents for efficient querying, and (2) retrieving complete document copies from the database based on search results. This segmentation allows the system to maintain high processing efficiency while eliminating information loss, as the complete documents are obtained from the database rather than attempting to recognize entire documents directly.
3Measurement precision
If machine recognition algorithms are refined to reduce errors, then recognition accuracy improves, but system complexity and computational resources increase
Solution Approach 1:
The patent extracts the complexity burden from the machine recognition algorithms by separating the recognition function from the retrieval function. Instead of relying on complex algorithms to achieve high accuracy across entire documents, the system extracts only the essential function of obtaining partial recognized text, which serves as a query input. The complex task of achieving complete and accurate document recognition is transferred to the search and retrieval process, which operates on the extracted partial information.
Data Source
AI summary
A system and method for capturing and recognizing at least a portion of a source document, whether written or audible, then searching for information, or other documents, that correspond to the captured and recognized portion of the source document. Various techniques for adding translation and/or searching are also disclosed. In some instances, an iterative machine learning process is applied to improve the performance of an aspect of the system.