Browse Prior Art Database

Baseline Drift Correction of Handwritten Text

IP.com Disclosure Number: IPCOM000045368D
Original Publication Date: 1983-Mar-01
Included in the Prior Art Database: 2005-Feb-06
Document File: 3 page(s) / 36K

Publishing Venue

IBM

Related People

Kim, J: AUTHOR

Abstract

For most people, handwriting goes up as writing progresses from left to right. When such writing is picked up by an electronic tablet for computer processing and displayed, it looks unnatural. If the hand writing has been formatted (word separation, text fill and spill etc.), baseline that is not horizontal looks even worse.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 54% of the total text.

Page 1 of 3

Baseline Drift Correction of Handwritten Text

For most people, handwriting goes up as writing progresses from left to right. When such writing is picked up by an electronic tablet for computer processing and displayed, it looks unnatural. If the hand writing has been formatted (word separation, text fill and spill etc.), baseline that is not horizontal looks even worse.

A method is described herein which can be used to correct drift in baseline such that a line of text lies on a horizontal line. This is an essential preprocess for any application that deals with handwritten text.

Baseline is a line which the characters without descenders sit on. Although baselines are straight and horizontal in printed text, they are neither straight nor horizontal in handwriting. Usually they are almost straight, but go up or down. This is called a linear drift. Sometimes they are curved, which is called a nonlinear drift. The linear drift correction will be described first, and then extended for nonlinear drift.

The first step in correcting baseline drift is to detect how much drift there is. Then a transformation can be made to correct the drift.

Linear Drift Detection. The discussion here is limited to one word at a time for the sake of simplicity. Treatment of a group of words at a time is discussed later.

A given word is divided into many smaller units. Each unit is a box of equal width and high enough to enclose all the points that lie within its width. Such boxes are uniformly spaced over the word (Fig. 1). The it is possible to analyze the relative position of the boxes and detect how they go up or down. Following is a simple-minded drift detection algorithm.
1. Find the total span of the word in x direction.
2. Divide the x span into smaller units of equal width.
3. For each x interval, draw a rectangular box that includes all

points that lie within the interval.
4. Discard the boxes with extreme height.
5. Among remaining boxes, find the box that has the average

height. This is the pivot box.
6. Using the center of the boxes, find dy/dx's for all boxes

from the pivot box.
7. Discard those dy/dx's that are extreme.
8. Get the average dy/dx from the remainder.

The average dy/dx is the amount of drift, which will be used for correction.

The performance depends on the number of boxes and their width, which in turn depends on the' size of writing, resolution, and sampling rate of the tablet. The number of boxes has to be large enough so that a meaningful number of boxes remain after discarding the extremes. The boxes have to be wide enough so that the enclosed points are meaningful, and also thin enough so that

1

Page 2 of 3

ascenders and descenders are isolated to a few box...