Browse Prior Art Database

Real-time Edge Detection Rotation of Check Images

IP.com Disclosure Number: IPCOM000110159D
Original Publication Date: 1992-Sep-01
Included in the Prior Art Database: 2005-Mar-25
Document File: 5 page(s) / 221K

Publishing Venue

IBM

Related People

Phan, NA: AUTHOR [+2]

Abstract

When a check passes through the transport of a Check Reader/Sorter (e.g., an IBM 3890), its image can be captured, compressed, and then sent to the host in real-time. However, the captured image will contain not only the check image, but also the transport image. In addition, due to imperfect check alignment systems, the check image will likely be skewed in either a clockwise or a counterclockwise direction.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 46% of the total text.

Real-time Edge Detection Rotation of Check Images

       When a check passes through the transport of a Check
Reader/Sorter (e.g., an IBM 3890), its image can be captured,
compressed, and then sent to the host in real-time.  However, the
captured image will contain not only the check image, but also the
transport image.  In addition, due to imperfect check alignment
systems, the check image will likely be skewed in either a clockwise
or a counterclockwise direction.

      The image edge detection/rotation algorithm described in this
article provides a solution to these problems.  The objective is to
identify, in real-time, the boundary around the check image.  Only
the image data within this boundary will then be kept for
compression; i.e., the compressed image will occupy less memory and
will also be skew-free.

      To be implemented in real-time, the algorithm will sample one
pel out of 16 pels, both horizontally and vertically, during the
capturing of image.  This is much faster compared with other methods
that process samples from the whole captured image.

      Each sampled pel will be assigned horizontal and vertical
coordinates, i.e., (x,y).  Any pel that has a value smaller than a
predetermined threshold T will be considered as inside the check
image and called a "light pel."  For simplicity, the figure on the
next page shows only the top and bottom light pels of each vertical
line; these pels roughly represent the top and bottom edges of the
check image.

      The image edge detection/rotation algorithm will first find the
initial boundary of the check image, then calculate the skew angle
based on either its top edge or its bottom edge (only pels within the
middle-half of an edge will be selected, since they seem to
characterize the slope most accurately).  Finally, the algorithm will
rotate the initial boundary around its center by the skew angle, the
rotation can be either clockwise or counterclockwise.

      The algorithm is described in three parts as follows:
      a.  Find the initial boundary ABCD.

      Starting from the right-most of the sampled image, look for the
first vertical line that contains at least one light pel, called x =
xLO .

      Starting from the left-most of the sampled image, look for the
first vertical line that contains at least one light pel, called x =
xHI .

                            (Image Omitted)

      Starting from the bottom of the sampled image, look for the
first horizontal line that contains at least one light pel, called
Y=YLO .

      Starting from the top of the sampled image, look for the first
horizontal line that contains at least one light pel, called Y=YHI .

      The four corners A (xLO, YLO), B(xLO, YHI), C (xHI, YHI),
D(xHI, YLO) make up the initial boundary of the check image. b.
Find the skew angle.

      Assume that the top edge, with a skew angle of Q, is chosen.
Since Q is gene...