Neural networks as for video signal processing

• US 5,161,014 A
• Filed: 03/12/1992
• Issued: 11/03/1992
• Est. Priority Date: 11/26/1990
• Status: Expired due to Term

• Current Assignee: RCA Thomson Licensing Corporation
• Original Assignee: RCA Thomson Licensing Corporation
• Inventors: Spence, Clay D., Pearson, John C.
• Primary Examiner(s): Kostak, Victor R.
• Application Number:
  US07/850,207
• Time in Patent Office: 236 Days
• Field of Search: 358/140, 358/160, 358/108, 358/141, 382/14, 382/15
• US Class Current: 348/443
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: H04N 7/012 Conversion between an interlaced and a progressive signal for graphics images G09G2310/0229
• CPC Class Codes: Y10S 706/908 Electronic or computer, internal or network, circuit

Abstract

A television signal processing apparatus includes at least one neural network for processing a signal representing an image. The neural network includes a plurality of perceptrons each of which includes circuitry for weighting a plurality of delayed representations of said signal, circuitry for providing sums of weighted signals provided by said weighting circuitry, and circuitry for processing said sums with a sigmoidal transfer function. The neural network also includes circuitry for combining output signals provided by ones of said perceptrons for providing a processed signal.

Claims

1. A television signal processing apparatus including at least one neural network for processing a video signal representing an image, said neural network including a plurality of perceptrons each of which includes means for weighting a plurality of delayed representations of said video signal, means for providing sums of weighted signals provided by said weighting means, and memory means addressed by said sums and programmed at respective memory address locations with values representing a sigmoidal function of address values of the form 1/(e^-Y +1) where Y corresponds to respective sums, said neural network also including means for combining output signals provided by ones of said perceptrons for providing a processed signal, and said apparatus including utilization means responsive to said processed signal.

2. The apparatus set forth in claim 1 wherein said television signal processing apparatus is a television receiver and said neural network is an interpolating filter responsive to said video signal for generating interpolated alternative video signal samples for use in one of video signal correction apparatus and interlace-to-noninterlace scan apparatus.

3. The apparatus set forth in claim 2 wherein said neural network includes;

delay means, responsive to said signal representing said image for concurrently providing a plurality of relatively delayed samples;

a first layer of N perceptrons (N an integer), each of said N perceptrons responsive to said plurality of relatively delayed samples;

at least one further perceptron responsive to signals provided by said first layer of N perceptrons, for providing said processed signal.

4. The apparatus set forth in claim 2 wherein said neural network includes;

delay means, responsive to said signal representing said image for concurrently providing a plurality of relatively delayed samples;

a first layer of N perceptrons (N an integer), each of said N perceptrons responsive to said plurality of relatively delayed samples;

a second layer of M perceptrons (M an integer), each of said M perceptrons responsive to signals provided by all of said plurality of N perceptrons; and

a third layer of Q perceptrons (Q an integer), each of said Q perceptrons responsive to signals provided by all of said plurality of M perceptrons, said third layer of perceptrons providing said processed signal.

5. The apparatus set forth in claim 1 wherein said neural network is a defect detector comprising:

delay means, responsive to said signal representing said image for concurrently providing a plurality of relatively delayed samples;

a first layer of N perceptrons (N an integer), each of said N perceptrons responsive to said plurality of relatively delayed samples;

at least one further perceptron responsive to signals provided by said first layer of N perceptrons, for providing said processed signal.

6. The apparatus set forth in claim 1 wherein said neural network is a defect detector comprising:

delay means, responsive to said signal representing said image for concurrently providing a plurality of relatively delayed samples;

a first layer of N perceptrons (N an integer), each of said N perceptrons responsive to said plurality of relatively delayed samples;

at least one further perceptron responsive to signals provided by said first layer of N perceptrons, for providing an intermediate processed signal;

further delay means, responsive to said intermediate processed signal for concurrently providing a plurality of relatively delayed samples representing said intermediate processed signal;

a first layer of M perceptrons (M an integer), each of said M perceptrons responsive to said plurality of relatively delayed samples representing said intermediate processed signal;

at least one further perceptron responsive to signals provided by said first layer of M perceptrons, for providing said processed signal.

7. The apparatus set forth in claim 4 wherein at least one successive layer of perceptrons is conditioned to operate at a lesser rate than a prior layer of perceptrons.

8. The apparatus set forth in claim 5 wherein said further delay means is coupled to said at least one further perceptron by a low pass filter.

9. A television signal processing apparatus including a detector for detecting defects in recovered video signals, said detector comprising a neural network responsive to recovered video signal, said neural network including a first plurality of perceptrons arranged to be operative on said video signal in one dimension (e.g. horizontal);

means for combining signals provided by said first plurality of perceptrons to provide an intermediate signal;

a second plurality of perceptrons arranged to be operative on said intermediate signal in two dimensions (e.g. horizontal and vertical); and

means for combining signals provided by said second plurality of perceptrons to form a signl indicative of video signal defects.

10. The apparatus set forth in claim 9 wherein said video signal occurs as samples at a predetermined sample rate;

said first plurality of perceptrons provide processed signals at a rate proportional to said predetermined rate; and

said neural network further includes a low pass filter for filtering said intermediate signal before application to said second plurality of perceptrons.

11. The apparatus set forth in claim 10 including means for conditioning said second plurality of perceptrons to provide processed signal at a lesser rate than said first plurality of perceptrons.

12. The apparatus set forth in claim 10 wherein said first plurality of perceptrons are arranged to simultaneously operate on a group of N samples, and wherein the first sample of each successive group of N samples is displaced at least two sample intervals from the first sample of the previous group of N samples (N an integer).

13. The apparatus set forth in claim 9 further including a second neural network responsive to said video signal for providing interpolated video signal;

and means responsive to said signal indicative of video signal defects for substituting said interpolated video signal for defects in said video signal.

14. A television signal processing apparatus including a neural network for processing recovered video signals, said neural network including a first plurality of perceptrons arranged to be operative on said video signal in one dimension (e.g., horizontal);

a second plurality of perceptrons arranged to be operative on subsampled signals provided by said first plurality of perceptrons and operative on signal in two dimensions (e.g., horizontal and vertical); and

means for combining signals provided by said second plurality of perceptrons to form a processed video signal.

15. The apparatus set forth in claim 14 wherein said first plurality of perceptrons are arranged to simultaneously operate on a group of N samples, and wherein the first sample of each successive group of N samples is displaced at least two sample intervals from the first sample of the previous group of N samples (N an integer).

16. The apparatus set forth in claim 14 wherein said neural network includes a further perception for combing signals from said first plurality of perceptions.

17. The apparatus set forth in claim 16 further including a low pass filter coupled between said further perception and said second plurality of perceptions.

18. The receiver set forth in claim 17 characterized in that said neural network is an interpolating filter for generating interpolated video signal sample values.

25. The apparatus set forth in claim 17 wherein said neural network includes a further perception for combing signals from said first plurality of perceptions.

19. A television signal processing apparatus including a neural network for processing recovered video signals, said neural network including a first plurality of perceptrons arranged to provide an intermediate signal;

a low pass filter for low pass filtering said intermediate signal;

a second plurality of perceptrons arranged to process lowpass filtered said intermediate signal and means for combining signals provided by said second plurality of perceptrons to form a processed video signal.

20. The apparatus set forth in claim 19 wherein said video signal occurs as samples at a predetermined sample rate;

said first plurality of perceptrons process samples at said predetermined rate; and

said second plurality of perceptrons process samples at a rate lower than said predetermined rate.

21. The apparatus set forth in claim 19 wherein said first plurality of perceptrons are arranged to simultaneously operate on a group of N samples, and wherein the first sample of each successive group of N samples is displaced at least two sample intervals from the first sample of the previous group of N samples (N an integer).

22. The receiver set forth in claim 19 wherein said first plurality of perceptrons is arranged to process said video signal in one dimension (e.g., horizontal) and said second plurality of perceptrons is arranged to process samples provided by said lowpass filter in two dimensions (e.g., horizontal and vertical).

23. The receiver set forth in claim 19 wherein each of said perceptrons of said first plurality includes means for weighting a plurality of delayed representations of said received video signal, means for providing sums of weighted signals provided by said weighting means, and means for processing said sums with a sigmoidal transfer function.

24. The receiver set forth in claim 19 characterized in that said neural network is an interpolating filter for generating interpolated video signal sample values.