Image processing method including image segmentation

• US 4,764,971 A
• Filed: 11/25/1985
• Issued: 08/16/1988
• Est. Priority Date: 11/25/1985
• Status: Expired due to Term

• Current Assignee: Eastman Kodak Company
• Original Assignee: Eastman Kodak Company
• Inventors: Sullivan, James R.
• Primary Examiner(s): Lall, Parshotam S.
• Assistant Examiner(s): MELNICK, STEVEN A
• Application Number:
  US06/802,113
• Time in Patent Office: 995 Days
• Field of Search: 364/554, 364/518, 364/521, 364/577, 340/738, 382/18, 382/51, 382/9, 382/52, 382/22, 382/50, 382/41, 375/25-28, 375/34, 358/135, 332/11 D
• US Class Current: 382/173
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling

Abstract

An image signal is segmented into regions of constant contrast or variance by applying a generalized-likelihood-ratio test to an image difference signal.

Claims

1. An image processing method comprising the steps of:

a. scanning an image to produce an image signal;

b. digitizing the image signal to produce a digital image having pixel amplitude values;

c. forming an image difference signal by taking differences between successive pixel amplitude values in the digital image;

d. segmenting in segmenting means the digital image into segments of pixel amplitude values having constant variance by applying a generalized likelihood ratio test (GLRT) to the image difference signal, to identify image segments, the GLRT testing for the hypothesis H.sub.φ

that a first set of image amplitude difference values from a Laplacian distribution having a variance σ

are statistically similar to a second set of image amplitude difference values having a variance σ

₀ to within a confidence level α

, where α

is an operator input value, the GLRT employing a generalized likelihood ratio (GLR) that is the ratio of maximum likelihood (ML) estimators for the variance σ

of a distribution having a Laplacian probability density function (pdf);

e. supplying said segments to image processing means;

f. processing in said image processing means the digital image employing said identified image segments to produce a processed digital image; and

g. supplying said processed digital image to an output device.

2. The image processing method claimed in claim 1, wherein said step of segmenting further includes the steps of:

a. estimating, in estimating means, the variance σ

₀ of pixel amplitude difference values in a first segment of said image difference signal using the ML estimator;

b. estimating, in estimating means, the variance σ

of the pixel amplitude difference values in a second segment of said image difference signal, including the last pixel amplitude difference value of said first segment and the next adjacent pixel amplitude difference value using the ML estimator;

c. applying said GLRT to σ and

σ

_0'"'"', if the hypothesis H.sub.φ

is true, adding the second segment to the first segment to generate a single segment, and repeating the steps of estimating, or if the hypothesis is false, identifying said first segment as an identified segment and beginning a new first segment with said next adjacent pixel amplitude difference value; and

d. repeating said steps of estimating and applying said GLRT until said image difference signal is completely segmented.

3. The image processing method claimed in claim 2, wherein said ML estimator is of the form:

##EQU16## where Δ

S_i is the ith pixel difference value and n is the number of pixel difference values in a segment.

4. The image processing method claimed in claim 3, wherein said GLR test is of the form:

reject the hypothesis H.sub.φ

whenever σ

/σ

₀ ≦

a or ≧

b where a and b are predetermined functions of the confidence level α

, defined by the following relationships;

##EQU17## where r is a dummy variable employed in the summation, and n is the number of pixel difference values in a segment.