Document Object Detection Using Morphological Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for identifying signature fields, seals, and other objects in documents rely heavily on manual intervention and are inefficient, especially when dealing with multiple languages and non-textual data, as they often require noise reduction and are limited by template specificity and language dependence.
Innovation Solution
A method and system that segment images based on visual attributes, perform morphological operations, and identify target objects using neighborhood information to automatically determine the presence or absence of signature fields, initials, notary seals, and other target objects, eliminating the need for manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If OCR technique and text matching approaches are used for identification, then text-based signature fields can be identified, but the system cannot handle documents in different languages and fails on non-textual objects like logos and seals
Solution Approach 1:
The patent replaces OCR-based text recognition with a deep learning-based image recognition system that processes visual features directly. This substitution enables the system to identify signature fields, seals, and logos without relying on textual content, making it language-agnostic and capable of handling non-textual objects while maintaining high identification accuracy through neural network-based pattern recognition
Solution Approach 2:
The patent creates a universal identification system that can handle multiple types of objects (signature fields, seals, logos) and documents in any language using a single deep learning model. This multi-functional approach eliminates the need for separate OCR pipelines for different languages and object types, achieving both versatility and reliability simultaneously
2Productivity
If conventional rule-based systems are used for document processing, then specified templates can be processed efficiently, but the system fails for new templates and requires manual intervention
Solution Approach 1:
The patent implements a dynamic identification system using deep learning that can adapt to new document templates automatically. Instead of static rule-based systems that require manual reconfiguration for each template, the neural network learns patterns from training data and generalizes to new templates, maintaining both processing speed and template flexibility through adaptive pattern recognition
3Measurement precision
If manual intervention is used for identifying signature fields and preparing documents for digital signature, then accuracy can be maintained, but the process becomes time-consuming and inefficient
Solution Approach 1:
The patent implements a self-service document processing system where the deep learning model automatically identifies signature fields, seals, and logos without human intervention. The system performs end-to-end processing including image segmentation, object detection, and field identification autonomously, achieving both high accuracy and rapid processing by eliminating manual steps while using automated neural network-based recognition
Data Source
AI summary
The present disclosure discloses a method and an object determination system for determining one or more target objects in an image. The image is segmented by the object detection system into one or more segments based on visual attributes in a first set. Morphological operations are performed on the one or more segments to obtain one or more morphed segments. One or more candidates of target objects are identified based on visual attributes in a second set corresponding to each one or more morphed segments. The object determination system identifies at least one of true positive and false positive from the one or more candidates which indicates presence or absence of the one or more target objects respectively, based on neighborhood information associated with the one or more candidates. The present disclosure facilitates in determining target objects in document automatically, thereby eliminating manual intervention in identifying target objects in the document.