Browse Prior Art Database

# High Speed Median Filtering for Gray Image Data

IP.com Disclosure Number: IPCOM000122474D
Original Publication Date: 1991-Dec-01
Included in the Prior Art Database: 2005-Apr-04
Document File: 4 page(s) / 98K

IBM

## Related People

Nakajima, M: AUTHOR [+1]

## Abstract

Disclosed is a high-speed median filtering algorithm for gray scale image data. The advantage of this algorithm is high-speed computation, expandability (independent from window size), and suitability for H/W implementation.

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

High Speed Median Filtering for Gray Image Data

Disclosed is a high-speed median filtering algorithm for
gray scale image data. The advantage of this algorithm is high-speed
computation, expandability (independent from window size), and
suitability for H/W implementation.

The median filter is a noise reduction filter for Gray Image
data.  The filtering operation is done for (n x n) window elements, n
is odd number and the target pixel is placed at the center of the n
x n window. The elements of the (n x n) window are sorted by its
value, and the median value is defined as the target pixel (see
Figure 1).

So this filtering operation requires (for the n X n window):
(n x n) times image memory access
O((n x n) log (n x n) .. (n x n x n x n))
computation for sorting on each pixel.

In this algorithm, sorting process is done by making histogram,
instead of comparing and swapping operation. The content of histogram
is inclemented by 1 when the element of window correspond to the
histogram entry (see Figure 2).

The median value is found by the following 3 registers.
LGR - Number of the members lower than median value
HGR - Number of the members higher than median value
MVR - Current median value
When LGR and HGR satisfy following expressions, the MVR is equal to
the median value.
LGR <= (n x n - 1)/2    (= 4)
HGR <= (n x n - 1)/2    (= 4)      - expression 1

When processing the next window, using a window register
(Figure 3), only n times Image Memory access are required. As Window
register keeps current window pixel data, window register shifts left
1 pixel on each row and shift in new n (=3) pixels.

The content of the histogram entry, which corresponds to the
shifted-out pixel, is decremente...