Document Field Detection via Template Feature Map Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image and text detection technologies in OCR systems are inadequate for directly identifying specific fields in documents, such as identity cards or driver licenses, as they require complex post-processing or customized models for each type of document, which is inefficient and not universally applicable.
Innovation Solution
A method and apparatus for processing images by acquiring a template image with defined regions of interest, determining feature maps for both the template and target images, and positioning regions of interest in the target image based on the similarity between the feature maps, allowing for customizable detection of fields in various document types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If simple image and text detection methods are used, then the detection process is fast and simple, but the system cannot satisfy the direct use requirement for semantically structured field definition output
Solution Approach 1:
The patent segments the detection task into two independent components: (1) a general text detection model that quickly locates all text regions, and (2) a field definition detection model that specifically identifies field boundaries and semantic structures. This segmentation allows the system to maintain fast processing while achieving precise field-level detection accuracy, as each component can be optimized independently for its specific function.
Solution Approach 2:
The patent introduces an intermediary component - the field definition detection model - that bridges the gap between simple text detection and complex semantic field extraction. This intermediary takes the detection results from the text detection model and transforms them into structured field definitions, enabling the system to satisfy both speed and precision requirements without requiring a completely complex end-to-end model.
2Measurement precision
If a complex post-processing or highly customized detection model is used, then the field detection accuracy is improved, but the device complexity and processing time increase
Solution Approach 1:
The patent divides the complex field detection task into two simpler, specialized models: a text detection model for locating text regions and a field definition detection model for identifying field structures. This segmentation reduces the complexity of each individual model compared to a single comprehensive model, while maintaining high field detection accuracy through the coordinated work of both specialized components.
Solution Approach 2:
The field definition detection model is designed with universal applicability to handle different types of fields (e.g., identity card fields, driver license fields, medical record fields) through a unified detection framework. This universality reduces the need for multiple highly customized models for different document types, thereby reducing overall system complexity while maintaining high detection accuracy across various field types.
3Measurement precision
If a customized detection model is trained for each document type, then the detection accuracy for that specific type is improved, but the adaptability to new document types decreases
Solution Approach 1:
The patent designs the field definition detection model with universal functionality that can handle multiple document types (identity cards, driver licenses, medical records, etc.) through a unified detection framework. The model uses configurable field definition templates that can be adjusted for different document types without requiring complete retraining, thereby maintaining high detection accuracy while achieving broad adaptability across various document categories.
Solution Approach 2:
The patent implements a dynamic field definition detection model that can adapt its detection parameters and field templates based on the input document type. The model dynamically adjusts its detection strategy and field definitions according to the specific document being processed, enabling it to maintain high accuracy across different document types without requiring static, document-specific models for each category.
4Measurement precision
If multiple separate detection models are trained for different fields, then the detection precision for each field is improved, but the processing time and computational resources increase
Solution Approach 1:
The patent merges the functionality of multiple separate field detection models into a single unified field definition detection model. This unified model simultaneously detects multiple field types (name, address, card number, etc.) in a single processing pass, thereby maintaining the detection precision of specialized models while eliminating the need for sequential processing of multiple separate models, thus reducing overall processing time and computational overhead.
Data Source
AI summary
A method and apparatus for processing an image, a device and a storage medium are provided. An implementation of the method includes: acquiring a template image, the template image including at least one region of interest; determining a first feature map corresponding to each region of interest in the template image; acquiring a target image; determining a second feature map of the target image; and determining at least one region of interest in the target image according to the first feature map and the second feature map.