Image content classification methods, systems and computer programs using texture patterns

Claim 2

An image content classification method comprising the steps of: repeatedly scanning an image comprising an array of image pixels, with at least one random neural network, each scan of the image with at least one random neural network corresponding to one of a plurality of texture patterns, wherein said repeatedly scanning step is preceded by the step of training said at least one random neural network to recognize said plurality of texture patterns, and wherein said training step comprises the steps of obtaining a plurality of samples of each of said plurality of texture patterns and applying said plurality of samples to said at least one random neural network; comparing a corresponding texture pattern to each of a plurality of image portions, each comprising at least one image pixel, for each of said plurality of scans of the image with at least one random neural network; and assigning to each image portion, a value corresponding to one of said plurality of texture patterns having highest coincidence.

Claim 3

A method according to claim 1 wherein said repeatedly scanning step comprises the steps of: providing a plurality of random neural networks corresponding to said plurality of texture patterns; and scanning each of said plurality of random neural networks once over said array of image pixels.

Claim 4

A method according to claim 1 wherein said repeatedly scanning step comprises the step of: providing a single random neural network; and scanning said single random neural network over said array of image pixels a plurality of times, corresponding to said plurality of image textures.

Claim 5

A method according to claim 1 wherein said assigning step comprises the step of selecting the image texture pattern having least mean squared error.

Claim 6

A method according to claim 1 wherein said random neural network comprises a bidirectionally interconnected neuron array, each neuron of which has a weight corresponding to an image texture pattern associated therewith.

Claim 7

A method according to claim 5 wherein each neuron of said bidirectionally connected array emits excitation and inhibition pulses.

Claim 8

A method according to claim 1 wherein said assigning step is followed by the step of: displaying an array of pixels representing the assigned values for the image portions.

Claim 9

A method according to claim 1 wherein said assigning step comprises the steps of: assigning to each image portion, a value corresponding to one of said plurality of texture patterns having highest coincidence, if each of said image patterns has a coincidence for the image portion which exceeds a threshold; and assigning an error value otherwise.

Claim 10

A method according to claim 1 wherein said obtaining step comprises the steps of: obtaining from an expert, an identification of said plurality of texture patterns; and obtaining from an expert, a plurality of samples for each of the identified texture patterns.

Claim 11

A method according to claim 1 wherein said image portion is a single image pixel.

Claim 12

An image content classification method comprising the steps of: at least one random neural network; means for repeatedly scanning an image comprising an array of image pixels, with at least one random neural network, each scan of the image with at least one random neural network corresponding to one of a plurality of texture patterns; means for training said at least one random neural network to recognize said plurality if texture patterns, wherein said training means comprises means for obtaining a plurality of samples of each of said plurality of texture patterns, and means, responsive to said obtaining means, for applying said plurality of samples to said at least one random neural network; means, responsive to said repeatedly scanning means, for comparing a corresponding texture pattern to each of a plurality of image portions, each comprising at least one image pixel, for each of said plurality of scans of the image with at least one random neural network; and means, responsive to said comparing means, for assigning to each image portion, a value corresponding to one of said plurality of texture patterns having highest coincidence.

Claim 13

An image content classification method comprising the steps of: at least one random neural network; means for repeatedly scanning an image comprising an array of image pixels, with at least one random neural network, each scan of the image with at least one random neural network corresponding to one of a plurality of texture patterns; means for training said at least one random neural network to recognize said plurality if texture patterns, wherein said training means comprises means for obtaining a plurality of samples of each of said plurality of texture patterns, and means, responsive to said obtaining means, for applying said plurality of samples to said at least one random neural network; means, responsive to said repeatedly scanning means, for comparing a corresponding texture pattern to each of a plurality of image portions, each comprising at least one image pixel, for each of said plurality of scans of the image with at least one random neural network; and means, responsive to said comparing means, for assigning to each image portion, a value corresponding to one of said plurality of texture patterns having highest coincidence.

Claim 14

A system according to claim 11: wherein said at least one random neural network comprises a plurality of random neural networks corresponding to said plurality of texture patterns; and wherein said means for repeatedly scanning an image comprises means for scanning each of said plurality of random neural networks once over said array of image pixels.

Claim 15

A system according to claim 11: wherein said at least one random neural network comprises a single random neural network; and wherein said means for repeatedly scanning an image comprises means for scanning said single random neural network over said array of image pixels a plurality of times, corresponding to said plurality of image textures.

Claim 16

A system according to claim 11 wherein said assigning means comprises means for selecting the image texture pattern having least mean squared error.

Claim 17

A system according to claim 11 wherein said random neural network comprises a bidirectionally interconnected neuron array, each neuron of which has a weight corresponding to an image texture pattern associated therewith.

Claim 18

A system according to claim 15 wherein each neuron of said bidirectionally connected array emits excitation and inhibition pulses.

Claim 19

A system according to claim 11 further comprising: means, responsive to said assigning means, for displaying an array of pixels representing the assigned values for the image portions.

Claim 20

A system according to claim 11 wherein said assigning means comprises: means for assigning to each image portion, a value corresponding to one of said plurality of texture patterns having highest coincidence, if each of said image patterns has a coincidence for the image portion which exceeds a threshold; and means for assigning an error value otherwise.

Claim 21

A system according to claim 11 wherein said obtaining means comprises: first means for obtaining from an expert, an identification of said plurality of texture patterns; and second means, responsive to said first means, for obtaining from an expert, a plurality of samples for each of the identified texture patterns.

Claim 22

A system according to claim 11 wherein said image portion is a single image pixel.

Claim 23

A computer program product for image content classification, said computer program product comprising: a computer-readable storage medium having computer-readable program code means embodied in said medium, said computer-readable program code means comprising: a computer-readable program code means for implementing at least one random neural network; computer-readable program code means for repeatedly scanning an image comprising an array of image pixels, with at least one random neural network, each scan of the image with at least one random neural network corresponding to one of a plurality of texture patterns; computer-readable program code means for training said at least one random neural network to recognize said plurality if texture patterns, wherein said training means comprises computer-readable program code means for obtaining a plurality of samples of each of said plurality of texture patterns, and computer-readable program code means, responsive to said obtaining means, for applying said plurality of samples to said at least one random neural network; computer-readable program code means, responsive to said repeatedly scanning means, for comparing a corresponding texture pattern to each of a plurality of image portions, each comprising at least one image pixel, for each of said plurality of scans of the image with at least one random neural network; and computer-readable program code means, responsive to said comparing means, for assigning to each image portion, a value corresponding to one of said plurality of texture patterns having highest coincidence.

Claim 24

A computer program product for image content classification, said computer program product comprising: a computer-readable storage medium having computer-readable program code means embodied in said medium, said computer-readable program code means comprising: a computer-readable program code means for implementing at least one random neural network; computer-readable program code means for repeatedly scanning an image comprising an array of image pixels, with at least one random neural network, each scan of the image with at least one random neural network corresponding to one of a plurality of texture patterns; computer-readable program code means for training said at least one random neural network to recognize said plurality if texture patterns, wherein said training means comprises computer-readable program code means for obtaining a plurality of samples of each of said plurality of texture patterns, and computer-readable program code means, responsive to said obtaining means, for applying said plurality of samples to said at least one random neural network; computer-readable program code means, responsive to said repeatedly scanning means, for comparing a corresponding texture pattern to each of a plurality of image portions, each comprising at least one image pixel, for each of said plurality of scans of the image with at least one random neural network; and computer-readable program code means, responsive to said comparing means, for assigning to each image portion, a value corresponding to one of said plurality of texture patterns having highest coincidence.

Claim 25

A computer program product according to claim 21: wherein said at least one random neural network comprises a plurality of random neural networks corresponding to said plurality of texture patterns; and wherein said means for repeatedly scanning an image comprises means for scanning each of said plurality of random neural networks once over said array of image pixels.

Claim 26

A computer program product according to claim 21: wherein said at least one random neural network comprises a single random neural network; and wherein said means for repeatedly scanning an image comprises means for scanning said single random neural network over said array of image pixels a plurality of times, corresponding to said plurality of image textures.

Claim 27

A computer program product according to claim 21 wherein said assigning means comprises means for selecting the image texture pattern having least mean squared error.

Claim 28

A computer program product according to claim 21 wherein said random neural network comprises a bidirectionally interconnected neuron array, each neuron of which has a weight corresponding to an image texture pattern associated therewith.

Claim 29

A computer program product according to claim 25 wherein each neuron of said bidirectionally connected array emits excitation and inhibition pulses.

Claim 30

A computer program product according to claim 21 further comprising: computer-readable program code means, responsive to said assigning means, for displaying on a display an array of pixels representing the assigned values for the image portions.

Claim 31

A computer program product according to claim 21 wherein said assigning means comprises: computer-readable program code means for assigning to each image portion, a value corresponding to one of said plurality of texture patterns having highest coincidence, if each of said image patterns has a coincidence for the image portion which exceeds a threshold; and computer-readable program code means for assigning an error value otherwise.

Claim 32

A computer program product according to claim 21 wherein said obtaining means comprises: first computer-readable program code means for obtaining from an expert, an identification of said plurality of texture patterns; and second computer-readable program code means, responsive to said first means, for obtaining from an expert, a plurality of samples for each of the identified texture patterns.

Claim 33

A computer program product according to claim 21 wherein said image portion is a single image pixel.