Browse Prior Art Database

N Bit General Neighborhood Function Processing Element Operable at Image Rate

IP.com Disclosure Number: IPCOM000100261D
Original Publication Date: 1990-Mar-01
Included in the Prior Art Database: 2005-Mar-15
Document File: 6 page(s) / 270K

Publishing Venue

IBM

Related People

Kimmel, MJ: AUTHOR

Abstract

Disclosed is programmable hardware for generating any neighborhood function of N spatially related "neighbors" over a binary image, an operation sometimes called feature extraction. The pixels which contribute to the "neighborhood" can have any spatial relationship in the image stream, up to some maximum separation. The scheme works in one clock time steady-state, irrespective of the number of neighbors (except the maximum N is fixed by the hardware) and the complexity of the neighborhood function. Hence, it is suitable for use in reconfigurable networks of parallel image processing elements of which the MITE is typical (1,2).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 24% of the total text.

N Bit General Neighborhood Function Processing Element Operable at Image Rate

       Disclosed is programmable hardware for generating any
neighborhood function of N spatially related "neighbors" over a
binary image, an operation sometimes called feature extraction.  The
pixels which contribute to the "neighborhood" can have any spatial
relationship in the image stream, up to some maximum separation.  The
scheme works in one clock time steady-state, irrespective of the
number of neighbors (except the maximum N is fixed by the hardware)
and the complexity of the neighborhood function. Hence, it is
suitable for use in reconfigurable networks of parallel image
processing elements of which the MITE is typical (1,2).

      See reference (3) for a complete description of the problem.
In that patent a different solution to the same problem is presented.
 The emphasis of that disclosure is on attaining a high packaging
density in VLSI by restricting the total amount of delay memory in
any one VLSI chip.  This invention utilizes commercially available
memory, and attains even greater generality in terms of the number of
neighborhood elements which can be handled in one neighborhood.

      A reference pixel is a pixel chosen as the origin of a
neighborhood, i.e. all neighbors are spatially related to the
reference pixel.  An N neighborhood is defined as a collection of N
pixels spatially related in any fashion to the reference pixel.  Each
possible spatial combination of N neighbors of the reference pixel is
a different neighborhood.  The N Bit General Neighborhood Function,
or N Bit GNF for short, is defined as a boolean function of the N
Bits encompassing a reference pixel and its N-1 neighborhood.  This
is also sometimes called a "morphic transform" except that usually
the morphic transform relates to a specific eight neighborhood, that
of the eight closest neighbors.  Feature extraction is another name
given to the process of evaluating a boolean function over an area of
an image, although sometimes only an existence within a region (not
an image of the function) is computed [4].  We use the terms feature
extraction and N Bit GNF interchangeably.

      See the figure for one neighborhood processing element. The
inputs and outputs are:
1.   Configuration Bus.  The host uses this bus to send configuration
data into the registers and memory of the neighborhood function
generating logic.  Its width depends upon the system bus chosen.
2.   Image In.  A one-bit serial stream of the image to be
transformed.  One new bit arrives every pixel (clock) time.
3.   Image Out.  A one-bit serial stream of the transformed image.
One new bit is generated every pixel (clock) time.

      The constituent parts are:
1.   Configuration Logic 1.  This moves initialization or
configuration data from the host system controller over the
CONFIGURATION_BUS into the address generator 4 and the NF lookup
table RAM 7.
2.   Input...