Image processing

Patent metadata

Field	Value
Publication number	US-10489634-B2
Application number	US-201715716220-A
Country	US
Kind code	B2
Filing date	Sep 26, 2017
Priority date	Sep 27, 2016
Publication date	Nov 26, 2019
Grant date	Nov 26, 2019

Abstract

A method including receiving an indication that a first classifier has identified that an image includes an object of a predetermined class of objects. Image data that relates to the image is processed using a second classifier with a first training state, which determines whether the image data includes the object of the predetermined class of objects. In response to the determining, data relating to the image data is transmitted to a remote system. Update data relating to the transmitted data is received from the remote system. The training state of the second classifier is updated to a second training state in response to the update data such that the second classifier with the second training state would make a different determination of whether future image data similar to the image data includes an object of the predetermined class of objects than the second classifier with the first training state.

First claim

What is claimed is:

1. A method comprising:
processing image data, that relates to an image that has been identified by a first classifier as comprising an object of a predetermined class of objects, using a second classifier with a first training state;
identifying, from the processing of the image data using the second classifier with the first training state, that the image data comprises the object of the predetermined class of objects;
in response to said identifying step, transmitting data relating to the image data to a remote system;
receiving update data from the remote system, the update data relating to the transmitted data, the update data being indicative that the transmitted data relates to a false positive identified by the remote system; and
updating the training state of the second classifier to a second training state in response to the update data such that the second classifier with the second training state would make a different determination of whether future image data similar to the image data comprises an object of the predetermined class of objects than the second classifier with the first training state.

2. The method according to claim 1 , wherein the updating the training state of the second classifier comprises updating the training state of the second classifier using false positive training data, the false positive training data comprising data derived from the image data.

3. The method according to claim 1 , wherein the first training state of the second classifier is at least partly based on a plurality of sets of false positive training data each derived from image data representing at least part of a respective image of a plurality of images.

4. The method according to claim 3 , comprising, in response to determining that adding a further set of false positive training data to the plurality of sets of false positive training data will increase the number of sets of false positive training data in the plurality of sets of false positive training data beyond a predetermined threshold:
discarding a set of false positive training data from the plurality of sets of false positive training data; and
adding a further set of false positive training data to the plurality of sets of false positive training data.

5. The method according to claim 4 , wherein the further set of false positive training data comprises data derived from the image data.

6. The method according to claim 1 , wherein the updating the training state of the second classifier comprises updating the training state of the second classifier using one or more sets of true positive training data each derived from image data representing at least part of a respective image of a plurality of images.

7. The method according to claim 1 , wherein the second classifier uses a linear classification model.

8. The method according to claim 7 , wherein the linear classification model comprises at least one of: a support vector machine, a two-neuron classifier, or a Fisher discriminant.

9. The method according to claim 1 , wherein the image data comprises feature vectors derived from at least part of the image.

10. The method according to claim 1 , comprising processing the image data using the first classifier to identify, using the first classifier, that the image comprises the object of the predetermined class of objects.

11. The method according to claim 1 , wherein the remote system comprises a third classifier, the method further comprising:
processing the transmitted data using the third classifier to attempt to identify an incorrect identification that the image data comprises the object of the predetermined class of objects by the second classifier; and
determining from the processing of the transmitted data using the third classifier that the second classifier has incorrectly identified that the image data comprises the object of the predetermined class of objects to generate the update data, the update data being indicative that the second classifier has incorrectly determined that the image data comprises the object of the predetermined class of objects.

12. The method according to claim 11 , wherein the third classifier uses a deep neural network.

13. The method according to claim 1 , wherein the predetermined class of objects is at least one of: human faces or other objects characteristic of a human being.

14. The method according to claim 1 , comprising determining that the image data satisfies at least one predetermined data assessment criterion before at least one of: the processing the image data using the second classifier with the first training state or the transmitting the data relating to the image data to the remote system.

15. The method according to claim 1 , wherein the image data is derived from video data.

16. A processor system configured to:
process image data, that relates to an image that has been identified by a first classifier as comprising an object of a predetermined class of objects, using a second classifier with a first training state;
identify, from the processing of the image data using the second classifier with the first training state, that the image data comprises the object of the predetermined class of objects;
in response to said identifying step, transmit data relating to the image data to a remote system;
receive update data from the remote system, the update data relating to the transmitted data, the update data being indicative that the transmitted data relates to a false positive identified by the remote system; and
update the training state of the second classifier to a second training state in response to the update data such that the second classifier with the second training state would make a different determination of whether future image data similar to the image data comprises an object of the predetermined class of objects than the second classifier with the first training state.

17. The processor system according to claim 16 , wherein the first training state of the second classifier is at least partly based on a plurality of sets of false positive training data each derived from image data representing at least part of a respective image of a plurality of images.

18. The processor system according to claim 17 , wherein the further set of false positive training data comprises data derived from the image data.

19. A remote system for processing data relating to image data representing at least part of an image,
a first classifier having identified that the image comprises an object of a predetermined class of objects,
the image data representing at least part of the image that has been identified by the first classifier as comprising an object of a predetermined class of objects having been processed using a second classifier with a first training state,
the second classifier with the first training state having identified, from the processing of the image data, that the image data comprises an object of the predetermined class of objects,
the remote system comprising:
a network interface to receive the data relating to the image data representing the at least part of the image from a computing device;
at least one processor; and
storage coupled to the at least one processor,
wherein the storage comprises:
computer program code configured to, when processed by the at least one processor, implement a third classifier, the third classifier being configured to:
process the received data to attempt to identify an incorrect determination that the image data comprises the object

Assignees

• Apical Ltd
• Univ Leicester
• Apical Limited And Univ Of Leicester

Inventors

• Romanenko Ilya
• Gorban Alexander
• TYUKIN Ivan

Classifications

• G06F18/211

  Selection of the most significant subset of features · CPC title

• G06F18/00 Primary

  Pattern recognition · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06N20/00

  Machine learning · CPC title