SPIKING NEURON NETWORK SENSORY PROCESSING APPARATUS AND METHODS

• US 20140016858A1
• Filed: 07/12/2012
• Published: 01/16/2014
• Est. Priority Date: 07/12/2012
• Status: Active Grant

• Current Assignee: Brain Corporation
• Original Assignee: Brain Corporation
• Inventors: Richert, Micah
• Granted Patent: US 8,977,582 B2
• Time in Patent Office: Days
• US Class Current: 382/156
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06N 3/10 Interfaces, programming languages or software development kits, e.g. for simulating neural networks
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

Apparatus and methods for detecting salient features. In one implementation, an image processing apparatus utilizes latency coding and a spiking neuron network to encode image brightness into spike latency. The spike latency is compared to a saliency window in order to detect early responding neurons. Salient features of the image are associated with the early responding neurons. A dedicated inhibitory neuron receives salient feature indication and provides inhibitory signal to the remaining neurons within the network. The inhibition signal reduces probability of responses by the remaining neurons thereby facilitating salient feature detection within the image by the network. Salient feature detection can be used for example for image compression, background removal and content distribution.

Claims

1. A computerized neuron-based network image processing apparatus comprising a storage medium, said storage medium comprising a plurality of executable instructions being configured to, when executed:

provide feed-forward stimulus associated with a first portion of an image to at least a first plurality of neurons and second plurality of neurons of said network;

provide another feed-forward stimulus associated with another portion of said image to at least third plurality of neurons of said network;

cause said first plurality of neurons to encode a first attribute of said first portion into a first plurality of pulse latencies relative to an image onset time;

cause said second plurality of neurons to encode a second attribute of said first portion into a second plurality of pulse latencies relative said onset time, said second attribute characterizing a different physical characteristic of said image than said first attribute;

determine an inhibition indication, based at least in part on one or more pulses of said first plurality and said second plurality of pulses, that are characterized by latencies that are within a latency window; and

based at least in part on said inhibition indication, prevent encoding of said another portion by said third plurality of neurons.

2. The apparatus of claim 1, wherein:

said latency window is determined relative said onset time; and

said one or more pulses are characterized by latency values being lower compared to other pulses of said first plurality and said second plurality of pulse latencies whose values are outside said latency window.

3. The apparatus of claim 1, wherein:

said first portion of said image comprises a plurality of pixels associated with a feature;

said first attribute comprises luminance of pixels within said plurality of pixels; and

said second attribute comprises velocity of pixels within said plurality of pixels.

4. The apparatus of claim 3, wherein:

said inhibition indication is configured based at least in part on at least two or more pulses within said first plurality and said second plurality, said at least two or more pulses each having a pulse latency within said window, said at least two or more pulses corresponding to at least two pixels of said plurality of pixels; and

said prevention of said encoding of said another portion by said third plurality of neurons is characterized by an absence of pulses associated with pixels of said image that are spatially outside of said portion.

5. The apparatus of claim 3, wherein said inhibition indication is effectuated at least by an inhibition trace, said inhibition trace being incremented responsive to an occurrence of a pulse latency of said first plurality of pulses and said second plurality of pulses being within said latency window.

6. The apparatus of claim 4, wherein

said at least two or more pulses comprise pulses within said first plurality;

and

said at least two pixels are characterized by a common luminance.

7. The apparatus of claim 6, wherein:

said inhibition indication is further configured based at least in part on another two or more pulses within said second plurality of pulses having a pulse latency within said window, said another two or more pulses corresponding to said two pixels of said plurality of pixels; and

said at least two pixels are characterized by a common rate of displacement.

8. The apparatus of claim 1, wherein said plurality of executable instructions are further configured to, when executed:

provide another feed-forward stimulus associated with a first portion of a second image to said first plurality of neurons, said second image preceding said image, said image and said second image comprising a representation of a feature;

cause said first plurality of neurons to encode said first attribute of said first portion of said second image into a third plurality of pulse latencies relative to an onset time associated with said second image;

said first portion of said image and said first portion of said second image having at least two pixels characterized by common value of said first attribute; and

wherein said inhibition indication is configured based at least in part on (i) at least one or more pulses within said first plurality and said second plurality of pulses; and

(ii) least another pulse associated with said third plurality of pulse latencies having latency within said window.

9. A computerized method of detection of one or more salient features of an image by a spiking neuron network, the method comprising:

providing feed-forward stimulus comprising a spectral parameter of said image to a first portion and a second portion of said network;

based at least in part on said providing said stimulus, causing generation of a plurality of pulses by said first portion, said plurality of pulses configured to encode said parameter into pulse latency;

generating an inhibition signal based at least in part on two or more pulses of said plurality of pulses being proximate one another within a time interval; and

based at least in part on said inhibition indication, suppressing responses to said stimulus by at least some neurons of said second portion.

10. The method of claim 9, wherein:

said providing said feed-forward stimulus is characterized by a stimulus duration, said stimulus duration being greater than said time interval; and

said suppressing responses is characterized by an absence of responses from said second portion during said stimulus duration.

11. The method of claim 9, wherein:

said providing said feed-forward stimulus is characterized by a stimulus duration, said stimulus duration being greater than said time interval; and

said suppressing responses is capable of reducing a number of responses from said second portion during said stimulus duration as compared to a number of responses that are generated by said second portion during said stimulus duration and responsive to said stimulus.

12. The method of claim 9, wherein:

said providing said feed-forward stimulus is characterized by a stimulus duration, said stimulus duration being greater than said time interval; and

said suppressing responses reduces a probability of response generation by neurons within said second portion during said stimulus duration and responsive to said stimulus.

13. The method of claim 9, wherein:

said providing said feed-forward stimulus causes generation of a burst of pulses by at least one neuron of said network; and

said suppressing responses suppresses at least a portion of pulses within said burst from being generated responsive to said stimulus.

14. The method of claim 9, wherein said suppressing responses is configured based at least in part on a number of said two or more pulses.

15. The method of claim 9, wherein;

said providing said feed-forward stimulus is characterized by an onset time and stimulus duration, said stimulus duration being greater than said time interval; and

said suppressing responses is configured based at least in part on a time interval between said onset time and generation of said two or more pulses.

16. A spiking neuron sensory processing system, comprising:

an encoder apparatus comprising;

a plurality of excitatory neurons configured to encode feed-forward sensory stimulus into a plurality of pulses; and

at least one inhibitory neuron configured to provide an inhibitory indication to one or more of said plurality of excitatory neurons over a one or more inhibitory connections;

wherein;

said inhibitory indication is based at least in part on two or more of said plurality of pulses being received by said at least one inhibitory neuron over one or more feed-forward connections; and

said inhibitory indication is configured to of prevent at least one of said plurality of excitatory neurons from generating, subsequent to said provision of said inhibitory indication, at least one pulse during a stimulus interval.

17. The system of claim 16, wherein:

said inhibitory indication is characterized by an inhibition trace, said inhibition trace configured to be;

incremented responsive to said receipt of said two or more of said plurality of pulses;

decayed with time subsequent to said increment; and

said preventing is based at least in part on a value of said trace exceeding an inhibitory threshold.

18. The system of claim 17, wherein:

said feed-forward stimulus is provided to said plurality of excitatory neurons over a stimulus interval;

said two or more of said plurality of pulses are received within a saliency interval, said saliency interval being greater than zero and smaller than one quarter of said stimulus interval; and

said time decay is effectuated throughout at least a portion of said stimulus interval.

19. The system of claim 17, wherein:

individual ones of the one or more inhibitory connections is characterized by a connection weight; and

said increment of said inhibition trace is based on increasing the connection weight.

20. The system of claim 16, wherein said inhibitory indication is capable of reducing number of the plurality of pulses associated with the feed-forward sensory stimulus thereby effectuating sensory stimulus compression.

21. The system of claim 16, wherein:

said feed-forward sensory stimulus comprises a background portion; and

said inhibitory indication is capable of preventing the background portion from being encoded into the plurality of pulses associated with the feed-forward sensory stimulus.

22. The system of claim 21, wherein

said background portion comprises uncorrelated sensory input component;

said feed-forward sensory stimulus further comprises coherent signal component;

the plurality of pulses constitute an encoded input; and

said inhibitory indication is capable of preventing the uncorrelated sensory input component from being encoded into the plurality of pulses, thereby effectuating an increased signal to noise ratio of the encoded input.