Top down preprocessor for a machine vision system

• US 5,870,493 A
• Filed: 03/17/1997
• Issued: 02/09/1999
• Est. Priority Date: 03/02/1994
• Status: Expired due to Fees

• Current Assignee: ERIM International, Inc. (Altarum Institute), Health and Human Services Department of United States of America, Human Services National Institutes of Health
• Original Assignee: ERIM International, Inc. (Altarum Institute), United States Department Of Health And Human Services
• Inventors: Vogl, Thomas P., Blackwell, Kim T., Alkon, Daniel L.
• Primary Examiner(s): CHANG, JON CARLTON
• Application Number:
  US08/819,142
• Time in Patent Office: 694 Days
• Field of Search: 382/295, 382/156, 382/195, 382/205, 382/288, 382/302, 382/276, 382/261, 382/190
• US Class Current: 382/195
• CPC Class Codes: G06V 10/454 Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]

Abstract

An image recognition and classification system includes a preprocessor in which a "top-down" method is used to extract features from an image; an associative learning neural network system, which groups the features into patterns and classifies the patterns: and a feedback mechanism which improves system performance by tuning preprocessor scale, feature detection, and feature selection.

Claims

1. A method for preprocessing an image for a machine vision system which receives signals generated from the image, the method comprising the steps of:

(1) top-down preprocessing the image, the step of top-down preprocessing the image including;

(a) dividing the image into subimages, wherein each subimage is contained in a receptive field of the machine vision system;

(b) applying processing operators to signals generated from each subimage to compute the centroid of each subimage;

(c) translating the subimage so that the centroid of each subimage is centered in its receptive field;

(d) applying a function to signals generated from each subimage to extract micro-features;

(e) determining whether a stopping criterion is fulfilled for each subimage; and

(f) repeating steps (a)-(e), using the subimage as the image in step (a) for each subimage determined not to fulfill the stopping criterion;

(2) adaptive filter preprocessing the image to extract micro-features; and

(3) feature binding to group the micro-features extracted using the top-down preprocessing and the micro-features extracted using the adaptive filter preprocessing to obtain macro-features.

2. The method of claim 1, wherein the micro-features comprise edge, texture, line end and curvature information.

3. The method of claim 1, wherein the function is a Gabor function.

4. The method of claim 1, wherein the stopping criterion is a function of an overall system performance and a change in system performance chances a scale of said receptive field.

5. The method of claim 1, wherein the stopping criterion is a function of a change in texture operators measured with respect to a scale across the image.

6. The method of claim 1, wherein the stopping criterion is a function of a change in texture operators measured with respect to a location across the image.

7. The method of claim 1, wherein the step of top-down preprocessing the image further includes:

(g) determining an overall system performance;

(h) determining whether said overall system performance is satisfactory; and

(i) if said overall system performance is determined not to be satisfactory, changing a scale of the function in step (d) and repeating steps (a)-(h).

8. The method of claim 7, wherein the stopping criterion is based on said overall system performance, and a change in said overall system performance changes a scale of said receptive field.

9. The method of claim 1, wherein the adaptive filter preprocessing comprises extracting the micro-features using a neural network.

10. The method of claim 1, wherein the adaptive filter preprocessing comprises extracting the micro-features using a multi-layer DYSTAL.

11. The method of claim 1 further comprising the step of:

calculating centroid coordinates of the macro-features and relative coordinates of the micro-features.

12. The method of claim 1, wherein the machine vision system comprises a preprocessor for performing the method, the method further comprising the steps of:

inputting a feedback signal for the preprocessor; and

modifying the steps of top-down preprocessing, adaptive filter preprocessing, and feature binding based on the feedback signal.

13. The method of claim 12, wherein the step of modifying the step of top-down preprocessing based on the feedback signal comprises increasing resolution of the receptive field.

14. The method of claim 1, the method further comprising the step of:

texture preprocessing the image to extract micro-features; and

wherein the step of feature binding further comprises grouping the micro-features extracted using the top-down preprocessing, the adaptive filter preprocessing, and the texture preprocessing to obtain macro-features.

15. A preprocessor for a machine vision system, the preprocessor comprising:

a top-down preprocessor comprising;

computing means for computing a centroid of an image;

translating means for translating the image about the centroid to provide a centered image;

extracting means for extracting micro-features from the centered image;

stopping means for stopping further processing of the image, wherein the stopping means provides a stopping signal if a stopping criterion is fulfilled; and

subdividing means for selectively subdividing the image into a plurality of subimages, wherein each one of the subimages is supplied to the computing means as the image, and wherein the subdividing means is responsive to the stopping signal to not perform any further subdivision of the image;

an adaptive filter preprocessor for processing the image to extract micro-features; and

a feature binder for grouping the micro-features extracted using the top-down preprocessing and the micro-features extracted using the adaptive filter preprocessing to obtain macro-features.

16. A method for preprocessing an image for a machine vision system which receives signals generated from the image, the machine vision system having a preprocessor for performing the method, the method comprising:

(1) top-down preprocessing the image, the step of top-down preprocessing the image including;

(a) extracting micro-features from the image;

(b) determining if a stopping criterion corresponding to the feature has been fulfilled; and

(c) if the stopping criterion has not been fulfilled, dividing the image into a plurality of subimages; and

(d) repeating steps (a)-(c) for each one of the plurality of subimages using each one of the plurality of subimages as the image in step (a);

adaptive filter preprocessing the image to extract micro-features; and

(2) feature binding to group the micro-features extracted using the top-down preprocessing to obtain macro-features;

(3) inputting a feedback signal for the preprocessor; and

(4) modifying the steps of top-down preprocessing and feature binding based on the feedback signal.

17. The method of claim 16, wherein the feedback signal increases the resolution for exacting micro-features from the image in the step of top-down preprocessing.