Text Line Normalization via Sequence Recognizer Height Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for estimating text height in document images face inaccuracies due to the presence of ascenders, descenders, capital letters, and numbers, which complicate the normalization process for optical character recognition (OCR), and often ignore contextual information within documents.
Innovation Solution
A machine learning model is employed to estimate text heights by calculating features from text line images, including adjacent text lines and contextual information, using a combination of lowercase and capital text height measures, and incorporating a conversion ratio to normalize text lines effectively for OCR.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods are used to estimate text height, then the process is simple, but the accuracy is poor due to ascenders, descenders, capital letters, and numbers
Solution Approach 1:
The text line is segmented into multiple vertical positions, and the text height is divided into multiple components (lowercase height, capital height, ascender height, descender height). Each component is estimated separately using the sequence recognizer, allowing for more accurate overall height estimation while handling the complexity of different character types.
Solution Approach 2:
The patent transitions from estimating a single text height value to estimating multiple height components at different vertical positions. This dimensional expansion allows the system to account for ascenders, descenders, and capital letters by treating height as a multi-dimensional property rather than a single scalar value.
2Measurement precision
If contextual information from adjacent text lines is incorporated, then the accuracy of text height estimation improves, but the computational complexity increases
Solution Approach 1:
The sequence recognizer is trained in advance on a large dataset of text lines with known height annotations. This preliminary training phase allows the model to learn contextual patterns and height relationships, so that during actual text processing, the system can quickly estimate heights using the pre-learned knowledge without extensive real-time computation.
Solution Approach 2:
The patent uses the text height of adjacent text lines as contextual information to inform the estimation of the current text line's height. By copying and comparing height patterns from neighboring lines, the system leverages contextual information to improve accuracy while avoiding the need for complex real-time analysis of all document content.
3Manufacturing precision
If multiple text height measures (lowercase and capital) are used, then the normalization accuracy improves, but the processing time increases
Solution Approach 1:
The system dynamically selects which height measures to compute based on the characteristics of the text line. The sequence recognizer determines the appropriate lowercase height, capital height, ascender height, and/or descender height to estimate based on the detected text features, avoiding unnecessary computations for text lines where certain measures are not applicable or needed.
Solution Approach 2:
The patent changes the parameter representation from a single text height value to multiple height components (lowercase, capital, ascender, descender). This parameter transformation allows for more precise normalization by capturing the nuanced vertical dimensions of different character types, while the sequence recognizer efficiently manages the computation of these multiple parameters.
Data Source
AI summary
A method for estimating text heights of text line images includes estimating a text height with a sequence recognizer. The method further includes normalizing a vertical dimension and/or position of text within a text line image based on the text height. The method may also further include calculating a feature of the text line image. In some examples, the sequence recognizer estimates the text height with a machine learning model.