Editable Image Documents via OCR Font and Color Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image documents with text are often not editable or require cumbersome image editing to modify text, limiting the ease and efficiency of text editing.
Innovation Solution
A computer-implemented method that uses optical character recognition (OCR) to predict the font and color of text in an image, forms editable text boxes, and inpaints the original image to create a modified document with editable text using a text editor.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If optical character recognition (OCR) is applied to extract text from image documents, then text becomes searchable and editable, but the text editing process becomes complex and requires specific image editing applications
Solution Approach 1:
The patent introduces an intermediary system that sits between the OCR extraction process and the text editing interface. This intermediary automatically extracts text, predicts formatting attributes (font, color, size), and reconstructs the document with editable text layers while preserving the original image background. This mediator eliminates the need for users to manually navigate complex image editing applications, as the system handles the conversion automatically.
Solution Approach 2:
The patent replaces the mechanical process of manual text extraction and formatting in image editing applications with an automated computer vision system. The system uses machine learning models to predict text attributes and automatically reconstructs editable text boxes, substituting the manual mechanical operations with intelligent automation that simplifies the user workflow.
2Ease of operation
If image documents are made editable with text, then text can be modified easily, but the original image content may be degraded or lost
Solution Approach 1:
The patent segments the document into distinct layers: the original image background layer and the editable text layer. By separating these components, the system allows text editing operations on the text layer without affecting the integrity of the original image layer. This segmentation enables independent modification of text while preserving the authentic visual context of the original document.
Solution Approach 2:
The patent creates a copy of the original image document and performs text extraction and editing operations on the copy rather than directly on the original. This copying approach ensures that the original image content remains intact and unchanged, while the edited version contains the modified text. The system uses the original image as a reference to maintain visual fidelity while enabling text modifications.
3Measurement precision
If text attributes like font and color are automatically predicted, then text boxes can be formed accurately, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary action by pre-training machine learning models on large datasets of text images during the system development phase. These pre-trained models have already learned to predict text attributes (font, color, size) with high accuracy. When processing actual documents, the system applies these pre-trained models directly, avoiding the need to perform complex training computations during document processing, thus reducing processing time while maintaining prediction accuracy.
Data Source
AI summary
Optical character recognition data for an image is used to generate a text box. The text box is formed to include text that has a colour and font determined based on analysis of the image. The colour may be determined using k-means clustering and the font determined using a trained image classification model. The text box may be located over the image at a location corresponding to the detected text in the image. The image may be inpainted at the location of the text box to remove the detected text from the image.