Non-linear color corrector having a neural network and using fuzzy membership values to correct color and a method thereof

• US 5,548,697 A
• Filed: 12/30/1994
• Issued: 08/20/1996
• Est. Priority Date: 12/30/1994
• Status: Expired due to Fees

• Current Assignee: Matsushita Electric Corporation Of America (Panasonic Holdings Corporation)
• Original Assignee: Panasonic Technologies, Inc. (Panasonic Holdings Corporation)
• Inventors: Zortea, Anthony E.
• Primary Examiner(s): Bayerl, Raymond J.
• Assistant Examiner(s): KATBAB, ABDOLLAH
• Application Number:
  US08/367,798
• Time in Patent Office: 599 Days
• Field of Search: 395/22, 395/61, 395/3, 348/650, 358/518, 382/159, 382/167, 382/165
• US Class Current: 706/2
• CPC Class Codes: H04N 9/646 for image enhancement, e.g. vertical detail restoration, cross-colour elimination, contour correction, chrominance trapping filters

Abstract

A color corrector for changing pixels in an image where the color corrector includes a neural fuzzy classifier to generate a membership value which defines a degree of membership of each pixel in a group of pixels to be transformed. A pixel color changer is also provided to transform the pixel according to its membership in the group of pixels to be changed. The color corrector can also include a pixel group classifier for identifying groups of pixels in the image to train the neural fuzzy classifier to generate the membership value.

Claims

1. A color corrector for transforming an image having pixels representing different color values in the image, said color corrector comprising:

means for identifying a grouping of pixels in the image to be transformed from a first group of color values towards a second color value;

neural fuzzy classifier means for only generating a membership data value for a pixel in the image defining a degree of membership of the pixel in the grouping of pixels, the neural fuzzy classifier means comprising a first neural network; and

transforming means for transforming the pixel by a specified amount in response to the membership data value toward the second color value where the transforming means is not a neural network.

2. The color corrector recited in claim 1, further including means for training the neural fuzzy classifier to generate the membership value, the training means including pixel classifier means for identifying pixel groups of related pixels in the image and means for generating weight data values in response to the pixel groups, said weight data values provided to the neural fuzzy classifier, the pixel fuzzy classifier means comprising a second neural network.

3. The color corrector recited in claim 2, wherein the image has do-nothing pixels not to be transformed and one of the weight data values represents a location of a do-nothing group of the do-nothing pixels in the image.

4. The color corrector recited in claim 2, wherein the pixel classifier means comprises a Kohonen Neural Network.

5. The color corrector recited in claim 1, wherein the neural fuzzy classifier comprises a feed-forward neural network.

6. A color corrector for transforming an image having pixels representing different color values in the image, said color corrector comprising:

means for identifying a grouping of pixels in the image to be transformed from a first grouping of color values towards a second color value and generating a change-class signal and a color change signal;

a fuzzy neural classifier coupled to the identifying means to receive the change-class signal, said fuzzy neural classifier only generating fuzzy membership signals, the neural fuzzy classifier means comprising a first neural network; and

a color corrector coupled to the neural classifier to receive the fuzzy membership signals and coupled to the identifying means to receive the color change signal, the color corrector for transforming the grouping of pixels by a specified amount towards the second color value where the color corrector is not a neural network.

7. The color corrector recited in claim 6, further including a pixel group classifier coupled to the fuzzy neural classifier to provide an example pixel signal, the pixel group classifier comprising a second neural network.

8. The color corrector recited in claim 7, wherein the pixel group classifier is a Kohonen Neural Network.

9. The color corrector recited in claim 6, wherein the fuzzy neural classifier is a feed-forward neural network.

10. The color corrector recited in claim 7, wherein the pixel group classifier identifies example pixel groups of the pixels in the image and wherein the example pixel groups are used to train the fuzzy neural classifier to produce the fuzzy membership signal specifying an amount to change a pixel in the image.

11. A method for transforming an image having pixels representing different color values in the image, said method comprising the steps of:

(1) identifying pixel groups of related pixels in the image using a first neural network applying unsupervised training;

(2) identifying a grouping of pixels in the image to be transformed from a first grouping of color values towards a second color value;

(3) generating, in response to the identified groups of related pixels, a membership data value for a pixel in the image defining a degree of membership of the pixel in the grouping of pixels using a second neural network applying supervised training; and

(4) transforming the pixel by a specified amount in response to the membership data value toward the second color value.

12. The method recited in claim 11, further including the step of identifying pixel groups of related pixels in the image, and wherein step (3) further includes generating the membership data in response to the grouping of pixels and the pixel to determine the degree of membership of the pixel in the grouping of pixels to be transformed.

13. A method for transforming an image having pixels representing different color values in the image, said method comprising the steps of:

(1) identifying a grouping of pixels in the image to be transformed from a first grouping of color values towards a second color value;

(2) generating a membership data value for a pixel in the image defining a degree of membership of the pixel in the grouping of pixels using a neural network; and

(3) transforming the pixel by a specified amount in response to the membership data value toward the second color value not using a neural network.