Text line segmentation method

Abstract

In a text line segmentation process, connected components (CCs) in document image are categorized into three subsets (normal, large, small) based on their sizes. The centroids of the normal size CCs are used to perform line detection using Hough transform. Among the detected candidate lines, those with line bounding box heights greater than a certain height are removed. For each normal size CC, if its bounding box does not overlap the bounting box of any line with an overlap area greater than a predefined fraction of the CC bounding box, a new line is added for this CC, which passes through the centroid of the CC and has an average slant angle. Each large size CCs are broken into two or more CCs. All CCs are then assigned to the nearest lines. A refinement method is also described, which can take any text line segmentation result and refine it.

Description

BACKGROUND OF THE INVENTIONField of the Invention[0001]This invention relates to document image processing, and in particular, it relates to a method for text line segmentation for document images.Description of Related Art[0002]Text line segmentation, i.e., segmenting lines of text from a document image (e.g. handwritten documents), is an important part of an intelligent character / word recognition (ICR / IWR) system. Many methods have been described, but it is difficult to obtain a general algorithm that can work well on a variety of writing samples having different slope, slant, inter-line connections, etc.[0003]In the field of offline intelligent word / character recognition, a conventional ICR / IWR system typically includes the following stages for processing an input text document image: text line segmentation; word / character segmentation; and a recognition module (usually CNN of kNN), to generate output word or character. Robust text line and word segmentation is a major bottleneck...

AI Technical Summary

Benefits of technology

The present invention introduces a better way to segment text lines in a ICR / IWR system. It includes a method to estimate the average height of characters and an improved technique for identifying text lines. These enhancements contribute to the overall accuracy and reliability of the system, allowing for more efficient and reliable processing of text data.

Problems solved by technology

Many methods have been described, but it is difficult to obtain a general algorithm that can work well on a variety of writing samples having different slope, slant, inter-line connections, etc.

Errors in text line segmentation lead to distorted input to the word / character recognition module and hence wrong output.

The RNN approach gives better recognition accuracy but suffers from very high computational cost and is currently difficult to deploy on a real time offline ICR / IWR system.

Images