Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Binary Thresholding of Check Images Using Gray Level Histograms

IP.com Disclosure Number: IPCOM000121509D
Original Publication Date: 1991-Sep-01
Included in the Prior Art Database: 2005-Apr-03
Document File: 4 page(s) / 135K

Publishing Venue

IBM

Related People

Narasimha, MS: AUTHOR [+2]

Abstract

This article describes a method of thresholding multilevel check images to binary or black/white images. Threshold levels are chosen by analyzing the gray level histograms of successive regions of the check image. This method has been found to give superior visual quality images than prior thresholding methods, especially on scenic checks. Equally important, this method provides images which are more compressible than those of prior techniques.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Binary Thresholding of Check Images Using Gray Level Histograms

      This article describes a method of thresholding
multilevel check images to binary or black/white images. Threshold
levels are chosen by analyzing the gray level histograms of
successive regions of the check image. This method has been found to
give superior visual quality images than prior thresholding methods,
especially on scenic checks.  Equally important, this method provides
images which are more compressible than those of prior techniques.

      Using gray level histograms for thresholding images is not new.
However, prior histogram methods have required that the histograms be
at least bimodal, so that a black level and a white level can be
chosen. The histogram analysis proposed in this article removes that
requirement.
THRESHOLDING BASICS

      A multilevel image f(x,y) is converted to a binary image by the
formula
      where b(x,y) is the binary value of the pel (black=0, white=1),
x and y are the pel coordinates, and t(x,y) is the threshold level
function.

      For non-scenic checks, the threshold level function t(x,y) can
be set to one level for the entire check, e.g., t(x,y) = 128. With
scenic checks, however, it is necessary to vary the threshold
function over the entire check in order to adapt to the changing
background level.  For best visual quality of the handwritten data
and for best compressibility, it is desirable to treat the scene as
background, and thus remove it from the image. In generating the
threshold level function t(x,y) then, it is necessary to identify
which gray levels within an image belong to the background, and
should therefore be discarded, and which belong to the data of
interest, such as the handwriting and machine printed characters, and
should therefore be kept.
PROPOSED THRESHOLDING METHOD

      The thresholding method we are proposing is an enhancement to
histogram thresholding techniques found in the literature. The first
step in the method is to divide a multilevel image into small windows
of m by n pels. These windows may be overlapped, which in general
will provide a better quality image. We have found that a window size
of m=n=32 works well. For each window, a histogram of the gray levels
is generated, an example of which is shown below. This histogram
assumes a 16-level (4-bit) image.

      From this bimodal histogram it is readily apparent that the
gray levels corresponding to black are those clustered about the
black peak (at 1), and those corresponding to the background are
those clustered about the white peak (at 9). Thus an appropriate
threshold level would fall between 1 and 9. The actual threshold used
is determined by a Look Up Table, which is indexed by the the
location of the black and white peaks. An example is shown below.

      Thus a black peak at 1 and white peak at 9 would provide a
threshold of T...