Method and apparatus for spawning specialist belief propagation networks based upon data classification

• US 8,891,859 B2
• Filed: 01/01/2014
• Issued: 11/18/2014
• Est. Priority Date: 09/02/2010
• Status: Active Grant

• Current Assignee: Golden Edge Holding Corporation
• Original Assignee: Edge Technologies, Inc
• Inventors: El Dokor, Tarek
• Primary Examiner(s): Mehta, Bhavesh
• Assistant Examiner(s): Harandi, Siamak
• Application Number:
  US14/145,945
• Publication Number:
  US 20140112577A1
• Time in Patent Office: 321 Days
• Field of Search: None
• US Class Current: 382/159
• CPC Class Codes: G06F 18/217 Validation; Performance evaluation; Active pattern learning techniques
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06F 18/24133 Distances to prototypes
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/776 Validation; Performance evaluation
• CPC Class Codes: G06V 20/10 Terrestrial scenes scenes under surveillance with static cameras G06V20/52; scenes perceived from the exterior of a vehicle G06V20/56; scenes perceived from the interior of a vehicle G06V20/59
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions
• CPC Class Codes: G06V 40/20 Movements or behaviour, e.g. gesture recognition recognition of facial expressions G06V40/16
• CPC Class Codes: H04N 5/33 Transforming infrared radiation cameras or camera modules for generating image signals from infrared radiation H04N23/20; circuitry of SSIS for transforming infrared radiation into image signals H04N25/20

Abstract

A method and apparatus for processing image data is provided. The method includes the steps of employing a main processing network for classifying one or more features of the image data, employing a monitor processing network for determining one or more confusing classifications of the image data, and spawning a specialist processing network to process image data associated with the one or more confusing classifications.

Claims

1. A method for processing image data, comprising the steps of:

employing a main belief propagation network for classifying one or more features of the image data;

employing a monitor belief propagation network for determining one or more classifications of the image data as belonging to one of a background, foreground or both; and

spawning a specialist belief propagation network to process image data associated with one or more of the one or more determined classifications.

2. The method of claim 1, wherein the main belief propagation network, monitor belief propagation network, and specialist belief propagation network comprise restricted Boltzmann Machines.

3. The method of claim 1, wherein the specialist belief propagation network is spawned by the monitor belief propagation network.

4. The method of claim 1, further comprising the steps of:

employing the monitor belief propagation network for determining one or more classifications of the image data by the specialist belief propagation network; and

spawning a second specialist belief propagation network to process image data associated with one or more of the one or more determined classifications of the image data processed by the specialist belief propagation network.

5. The method of claim 4 wherein the second specialist belief propagation network further spawns a third or more specialist belief propagation networks.

6. The method of claim 1, wherein image data processed by the specialist belief propagation-network is provided to the main belief propagation network directly or via analysis and updates by the monitor belief propagation network.

7. The method of claim 1, wherein the monitor belief propagation network employs a volatility index for determining one or more classifications of the image data as belonging to one of the background, foreground or both.

8. The method of claim 1, wherein a single (x,y) pixel location may comprise classifications of multiple depth (z) objects.

9. The method of claim 7, wherein one or more of a foreground object, an occluded object and a background object may occupy a single (x,y) pixel location.

10. The method of claim 1, wherein the classifications comprise one or more scene objects, one or more salient scene features being used to train one or more of the main belief propagation networks and specialist belief propagation network.

11. The method of claim 1, wherein the step of classifying the one or more features of the image data further comprises the steps of:

classifying one or more scene exposure settings by the main belief propagation network according to one of scene exposure and scene quality;

spawning one or more specialist belief propagation networks, each adapted to process image data related to one or more of the one or more scene exposure setting classifications;

wherein each of the one or more specialist belief propagation networks is adapted to modify one or more exposure settings associated with the associated classification of exposure settings, thereby improving the exposure conditions thereof.

12. The method of claim 11, wherein the specialist belief propagation network is also adapted to modify one or more of chroma settings, saturation settings, color relevant settings and texture relevant settings.

13. The method of claim 1, wherein the classifications allow for the performance of disparity space decomposition via disparity decomposition metrics that change thresholds adaptively with disparity values.

14. The method of claim 13 wherein the performance of disparity space decomposition further comprises the step of extraction of “

energy nodules”

that are used in training one or more of the specialist belief propagation networks on various scene settings.

15. The method of claim 1, further comprising the step of computing a volatility index associated with the computation of a segment'"'"'s disparity in order to at least in part classify a segment as belong to a background, foreground or both in accordance with training one or more specialist belief propagation networks in accordance with one or more scene analysis and stability metrics.

16. The method of claim 1, wherein a more comprehensive topology is employed comprising a plurality of subnets branching from the main subnet to deal with various modalities including one or more of gesture and facial expression.

17. The method of claim 16, wherein one or more additional specialist subnets are spawned in the presence of an altogether different modality, unrelated to either facial or gesture expression recognition.

18. The method of claim 1, wherein the image data comprises data from a pair of stereo cameras sensitive to visible light, and data from a pair of stereo cameras sensitive to infrared light.

19. The method of claim 18, further comprising the step of determining whether to employ image data from the pair of stereo cameras sensitive to visible light or the pair of stereo cameras sensitive to infrared light in accordance with ambient lighting conditions.

20. The method of claim 7, wherein image data with a high volatility is determined to be part of the foreground.