Teachable pattern scoring method

• US 9,152,884 B2
• Filed: 06/05/2012
• Issued: 10/06/2015
• Est. Priority Date: 06/05/2012
• Status: Active Grant

• Current Assignee: Leica Microsystems Gmbh (Danaher Corporation)
• Original Assignee: DRVision Technologies, LLC (Danaher Corporation)
• Inventors: Lee, Shih-Jong J., Huang, Chi-Chou
• Primary Examiner(s): Le, Vu
• Assistant Examiner(s): WOLDEMARIAM, AKLILU K
• Application Number:
  US13/507,115
• Publication Number:
  US 20130322741A1
• Time in Patent Office: 1,218 Days
• Field of Search: 382/155, 382/159, 382/171, 382/173, 382/180, 382/224, 382/240
• US Class Current: 1/1
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/24137 Distances to cluster centroïds
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 20/695 Preprocessing, e.g. image segmentation
• CPC Class Codes: G06V 20/698 Matching; Classification

Abstract

A computerized teachable pattern scoring method receives a teaching image and region pattern labels. A region segmentation is performed using the teaching image to generate regions of interest output. A feature measurement is performed using the teaching image and the regions of interest to generate region features output. A pattern score learning is performed using the region features and the region pattern labels to generate pattern score recipe output. A computerized region classification method using the region features and the pattern score recipe to generate pattern scores output. A region classification is performed using the pattern scores and region features to generate region class output.

Claims

1. A computerized teachable pattern scoring method for segmented object region comprising the steps of:

a) Input a teaching image;

b) Perform region segmentation using the teaching image to generate regions of interest output;

c) Perform region feature measurement using the teaching image and the regions of interest to generate region features output wherein the region features are generated for each region of interest and a region of interest has a same region feature value;

d) Input region pattern labels;

e) Perform pattern score learning using the region features and the region pattern labels to generate region based pattern score recipe output wherein the pattern score recipe contains instructions for a computer to generate a pattern score wherein the pattern score is a distance metric with a normalization factor and wherein a region of interest has a same pattern score value.

2. The computerized teachable pattern scoring method of claim 1 wherein the region pattern labels is generated by the following steps:

a) Perform pattern type specification by user to generate pattern types;

b) Perform region labeling by user to generate region pattern labels.

3. The computerized teachable pattern scoring method of claim 1 wherein the pattern score learning comprises the steps of:

a) Perform pattern scoring model learning using the region features and the region pattern labels to generate pattern scoring model output;

b) Perform pattern scoring model storage using the pattern scoring model to generate pattern score recipe output.

4. The pattern score learning of claim 3 further comprises a feature selection step using the region pattern labels and region features to generate selected features output.

5. The pattern score learning of claim 3 wherein the pattern scoring model is selected from a set consists of distance to center model, multiple cluster centers model and kernel based model.

6. A computerized update teaching of a teachable pattern scoring method for segmented object region comprising the steps of:

a) Input a validation image;

b) Perform region segmentation using the validation image to generate regions of interest output;

c) Perform region feature measurement using the validation image and the regions of interest to generate region features output wherein the region features are generated for each region of interest and a region of interest has a same region feature value;

d) Input pattern score recipe;

e) Perform pattern scoring using the region features and the pattern score recipe to generate region based pattern scores output wherein a region of interest has a same pattern score value;

f) Perform additional pattern labeling using the region based pattern scores to generate additional region pattern labels output;

g) Perform pattern score update learning using the region features, the additional region pattern labels and the pattern score recipe to generate updated pattern score recipe output wherein the updated pattern score recipe contains instructions for a computer to generate a pattern score wherein the pattern score is a distance metric with a normalization factor.

7. The computerized update teaching method of claim 6 wherein the additional region pattern labels is performed using a computer GUI wherein regions of interest are displayed and user can perform labeling by simple clicking at regions where pattern scores are incorrect.

8. The computerized update teaching method of claim 6 wherein the pattern score update learning comprises the steps of:

a) Perform pattern scoring model update learning using region features, additional region pattern labels and the pattern score recipe to generate updated pattern scoring model output;

b) Perform pattern scoring model update storage step using the updated pattern scoring model to generate updated pattern score recipe output.

9. The pattern score learning of claim 8 wherein the pattern scoring model is selected from a set consists of distance to center model, multiple cluster centers model and kernel based model.

10. A computerized region classification method based on teachable pattern scoring comprising the steps of:

a) Input at least one application image;

b) Perform region segmentation using the application image to generate regions of interest output;

c) Perform region feature measurement using the application image and the regions of interest to generate region features output wherein the region features are generated for each region of interest and a region of interest has a same region feature value;

d) Input pattern score recipe wherein the pattern score recipe contains instructions for a computer to generate a pattern score;

e) Perform pattern scoring using the region features and the pattern score recipe to generate region based pattern scores output wherein the region based pattern scores are distance metrics with normalization factors and wherein a region of interest has a same pattern score value;

f) Perform region classification using the region based pattern scores and region features to generate region class output wherein a region of interest has a same pattern class value.

11. The region classification method of claim 10 wherein the region segmentation method is a learnable object segmentation.

12. The region classification method of claim 10 wherein the region features include morphological features, intensity features and texture and contextual features.

13. The region classification method of claim 10 wherein the pattern scoring generates pattern scores using distance metric to different pattern types.

14. The region classification method of claim 10 wherein the region classification using decision tree classifier.

15. The region classification method of claim 13 wherein the distance metric uses Mahalanobis distance.