Entity Matching via Deep Learning Vision Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for updating entity listings in geographic information systems, such as those using optical character recognition (OCR), are unreliable due to issues like occluded views, blurring, and difficult-to-transcribe signage, leading to inaccurate transcription of text from images.
Innovation Solution
A computer-implemented method that identifies entities in images and determines candidate entity profiles using a machine learning model, specifically a deep convolutional neural network (CNN) and long short-term memory (LSTM) network, to generate match scores between images and entity profiles without explicit text transcription, allowing for accurate entity matching and directory updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical character recognition (OCR) is used to transcribe text from images, then entity information can be extracted from images, but the transcription accuracy deteriorates due to occluded views, blurring, and difficult-to-transcribe signage
Solution Approach 1:
The patent replaces the mechanical/optical OCR system with a machine learning-based vision system. Instead of using traditional OCR that reads text character-by-character, the system uses a machine learning model trained to recognize entity information directly from images, substituting the mechanical text recognition process with an intelligent pattern recognition approach that can handle complex visual patterns without relying on explicit text transcription
Solution Approach 2:
The patent introduces an intermediary machine learning model that acts as a mediator between the image and the entity information extraction process. This intermediary layer processes the image data through learned features and patterns, translating visual information into structured entity data without requiring direct text transcription, thereby avoiding the inaccuracies of traditional OCR
2Measurement precision
If manual inputting of information into entity profiles is used, then information can be accurately entered, but the productivity and efficiency of updating entity listings deteriorates
Solution Approach 1:
The patent implements self-service by enabling the system to automatically extract and populate entity information from images without requiring manual intervention. The machine learning model autonomously processes images, identifies entities, extracts relevant information, and updates entity profiles automatically, allowing the system to serve itself rather than requiring human operators to manually enter data
Solution Approach 2:
The patent applies preliminary action by pre-processing and extracting entity information from images before the actual entity profile updating process. The machine learning model performs preliminary extraction of features, patterns, and potential entity data from images in advance, preparing the information so that it can be directly used to update entity profiles without requiring subsequent manual verification or entry
Data Source
AI summary
Systems and methods of identifying entities are disclosed. In particular, one or more images that depict an entity can be identified from a plurality of images. One or more candidate entity profiles can be determined from an entity directory based at least in part on the one or more images that depict the entity. The one or more images that depict the entity and the one or more candidate entity profiles can be provided as input to a machine learning model. One or more outputs of the machine learning model can be generated. Each output can include a match score associated with an image that depicts the entity and at least one candidate entity profile. The entity directory can be updated based at least in part on the one or more generated outputs of the machine learning model.