HandySeg 0.3.0

A package for line and word segmentation of handwritten scanned text

HandySeg package

1. Line Segmentation from Handwritten Text

Overview

Seam Carving and Horizontal Projection Profiles (HPP) are used to segment lines of handwritten text from an image. Initially, HPP is used to detect lines, ensuring the seam's path between lines does not shift outside the boundary, maintaining alignment based on low energy.

Figure A: Horizontal projection profile of handwritten text image

Figure B: Handwritten text image with crossed out lines

Energy calculation is performed, and an algorithm is applied to find the low path energies between lines, ensuring paths do not extend beyond the identified rows. This maintains the integrity of the seam's path.

Figure C: The lowest path energies between each two lines

Figure D: Segmented handwritten line text

2. Word Segmentation from Segmented Lines

Overview

First, a Close Morphological operation is applied to the segmented line using a kernel size dependent on both the width and length of the segmented line.

Figure E: The binary image after applying the Closing Morphological operation

Next, contours of the preprocessed segmented line image are found.

Figure F: Drawn a rectangle around each word

Figure G: Segmented word

Limitations

Line Segmentation:

• Does not work well on images with objects other than handwritten text.
• Not effective for lines with a large width.

Word Segmentation:

• Ineffective for words with excessive space between characters.
• Not suitable for words with minimal space between them.