Touching Character Segmentation via Pixel Density Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting touching characters in text outputs from videos and images are inefficient due to noise interference and incorrect segmentation, leading to low recognition accuracy, especially for text-rich videos where characters are not properly segmented.
Innovation Solution
A method and system that determine the aspect ratio of components, investigate geometric orientations, compute pixel densities, and segment touching characters using a candidate cut column based on pixel density and length, to accurately separate characters with a referential boundary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing segmentation methods are used for touching characters, then the process is simple, but the segmentation accuracy is low due to incorrect candidate cutting positions
Solution Approach 1:
The patent segments the text line into multiple columns based on pixel density analysis. It divides the touching character region into candidate cut columns by analyzing pixel density distribution across different positions, then selects optimal cut positions to separate individual characters. This segmentation approach directly addresses the accuracy problem by breaking down the complex touching character recognition into manageable column-based units.
Solution Approach 2:
The patent introduces a new dimension of analysis by computing pixel density not just horizontally but also vertically across multiple lines. It analyzes the relationship between horizontal position and vertical pixel density to identify candidate cut columns, transforming the problem from simple horizontal segmentation to a two-dimensional density analysis that more accurately captures character boundaries.
2Measurement precision
If existing methods assume equal character width, then the processing is straightforward, but the candidate cutting positions are incorrect leading to over-segmentation
Solution Approach 1:
The patent applies local quality analysis by computing pixel density for each specific column position rather than assuming uniform character width. It identifies regions with high pixel density as potential character bodies and low density regions as spacing, allowing each local region to be analyzed according to its actual pixel distribution characteristics. This enables accurate cutting position identification even when characters have varying widths.
Solution Approach 2:
The patent changes the parameter from fixed character width assumption to dynamic pixel density measurement. By computing actual pixel density values for each column and using these values to determine candidate cut positions, the method adapts to varying character widths and spacing in different text formats, maintaining high accuracy across diverse text styles and fonts.
3Reliability
If existing segmentation methods are applied to text-rich videos, then the processing speed is maintained, but the recognition accuracy deteriorates due to noise and improper segmentation
Solution Approach 1:
The patent extracts the essential feature of pixel density from the noisy video text images and uses this extracted feature for segmentation. By focusing on the pixel density distribution pattern rather than the raw noisy pixel values, the method separates the useful segmentation information from the harmful noise, enabling accurate character separation even in text-rich video environments with various noise types.
Solution Approach 2:
The patent implements a feedback mechanism by evaluating the pixel density relationship between adjacent columns to confirm candidate cut positions. It checks whether the pixel density pattern supports a proposed cut position by analyzing the density values and their relationships, providing feedback validation that reduces false segmentation caused by noise while maintaining processing efficiency for video text.
Data Source
AI summary
The present disclosure relates to a system and a method for detection of touching characters in a media, characterized by segmentation of adjoining character spaces. In the very first step, an aspect ratio is calculated for each connected component. A candidate touching position of each character is determined by calculating a threshold aspect ratio for each character. Further, a candidate cut column is determined based on a relation between column pixel densities and corresponding length thereof the column in order to segment the touching characters at the candidate cut column.