Semi-supervised method for training multiple pattern recognition and registration tool models

• US 9,659,236 B2
• Filed: 11/25/2015
• Issued: 05/23/2017
• Est. Priority Date: 06/28/2013
• Status: Active Grant

• Current Assignee: Cognex Corporation
• Original Assignee: Cognex Corporation
• Inventors: Michael, David J., Barker, Simon
• Primary Examiner(s): Liu, Li
• Application Number:
  US14/952,883
• Publication Number:
  US 20160155022A1
• Time in Patent Office: 545 Days
• CPC Class Codes: G06F 18/2155 characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
• CPC Class Codes: G06F 18/217 Validation; Performance evaluation; Active pattern learning techniques
• CPC Class Codes: G06F 18/2193 based on specific statistical tests
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 10/772 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries

Abstract

A system and method for training multiple pattern recognition and registration models commences with a first pattern model. The model is trained from multiple images. Composite models can be used to improve robustness or model small differences in appearance of a target region. Composite models combine data from noisy training images showing instances of underlying patterns to build a single model. A pattern recognition and registration model is generated that spans the entire range of appearances of the target pattern in the set of training images. The set of pattern models can be implemented as either separate instances of pattern finding models or as a pattern multi-model. The underlying models can be standard pattern finding models or pattern finding composite models, or a combination of both.

Claims

1. A method for training a pattern recognition and registration multi-model, the method comprising the steps of:

providing at least one initial training image having a region specifying a pattern to be trained, the at least one initial training image being provided from a database containing a plurality of training images;

training a first pattern model from the at least one initial training image and the region and adding the first pattern model to an output multi-model;

iterating over the remaining training images of the plurality of training images so that for each training image,

(i) an additional pattern model may be trained,

(ii) a metric for the combination of the first and additional model may be scored over the remaining training images in the database;

adding the high scoring additional pattern model or models to the output multi-model.

2. The method as set forth in claim 1 wherein the training of the first pattern model is performed according to a first set of pattern recognition and registration training parameters and the training of the additional pattern model is performed according to a second set of pattern recognition and registration training parameters.

3. The method as set forth in claim 1 wherein the first and/or the additional pattern model comprises a composite model.

4. The method as set forth in claim 1 wherein the additional pattern model is trained by relaxing the degrees-of-freedom when generating candidate regions using the first model to train, and then tightening the degrees-of-freedom when the second pattern model is run, to determine if a metric improves.

5. The method as set forth in claim 4 wherein the metric comprises the score or the number of instances of features.

6. The method as set forth in claim 4 wherein the additional pattern model is trained from the initial training image and specified region but with different training parameters so that the additional pattern model will be more likely to find distorted, noisy or in some way modified instances of the original pattern;

proposing candidate regions using a second pattern model, for use in training additional pattern models.

7. The method as set forth in claim 1 wherein a pattern origin is an additional input to pattern model training.

8. The method set forth in claim 1 wherein the high scoring additional pattern model is first presented to a user to accept, reject or rescore, before possible addition to the output multi-model.

9. The method set forth in claim 1 wherein a user may modify the additional pattern origin before its addition to the output multi-model.

10. The method set forth in claim 1 wherein the training process for the additional pattern model is repeatedly iteratively, thus allowing the multi-model to expand to contain greater than 2 pattern models.

11. The method set forth in claim 10 wherein a stopping condition is applied;

the stopping condition being derived from the relative improvement in the metric.

12. The method set forth in claim 1 but where supplied ground truths for the each of the plurality of images in the training image database may be used to accept or reject and/or correct the poses of candidate models for addition to the multi-model.