Method and apparatus for adaptive color scanning/printing data correction employing neural networks

• US 5,748,329 A
• Filed: 03/08/1996
• Issued: 05/05/1998
• Est. Priority Date: 03/08/1996
• Status: Expired due to Term

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Chang, Gao-Wei
• Primary Examiner(s): FRAHM, ERIC S
• Application Number:
  US08/612,574
• Time in Patent Office: 788 Days
• Field of Search: 358/298, 358/501, 358/504, 358/518, 358/520, 358/530, 358/534, 382/162, 382/167, 395/21-24
• US Class Current: 382/167
• CPC Class Codes: H04N 1/6033 using test pattern analysis H04N1/6055 takes precedence

Abstract

A method and apparatus for adaptive color scanning/printing data correction to effectively adjust to an arbitrary combination of color image scanning input and printing output devices for reproducing color image at optimized resemblance. An improved back propagation algorithm is employed to reduce the learning error and accelerate the process of learning procedure by increasing the rate of convergence. A method of characteristics extracting functionalization is utilized to urge the successful convergence of the learning procedure of the neural network, and reduce the color discrepancy and accelerate the process of learning convergence. An enhanced grey-scale balancing scheme is also utilized to extract the grey component of a learning sample under a predetermined condition and re fetch again to the neural network for accelerated convergence of the learning behavior. An apparatus for performing the color data correction includes an integrated neural network color processor that utilizes a small number of neural elements to reduce the complexity of the color data processing and computational complication.

Claims

1. A method for adaptive color scanning/printing data correction, wherein a color scanning/printing system includes a color image inputting device, a neural network, and a color image outputting device, said inputting device being connected to said neural network, and said neural network being connected to said outputting device, said method comprising the steps of:

providing pre-selected test color image patterns to said color image outputting device for outputting a first output image;

inputting said first output image to said inputting device to obtain color data of said first output image, and processing a learning sample generated with said color data by said neural network;

said neural network subsequently taking the average of the data values of all the color picture elements within a neighborhood in the predetermined vicinity of each input color picture element, and taking the average data value of all the color picture elements within the corresponding neighborhood of the corresponding output color picture element, so as to constitute a learning sample pair including the averaged color data values of input and output gamuts respectively; and

said neural network processing said generated learning sample pair to perform a neural network learning procedure.

2. The method of claim 1 for adaptive color scanning/printing data correction, wherein said neural network performs a back-propagation balancing algorithm for a learning procedure until the learning error reaches below a predetermined value, wherein said back-propagation algorithm is defined:

for output layer k of said neural network,

space="preserve" listing-type="equation">Δ

W.sub.kj =η

.δ

.sub.k.Oj

space="preserve" listing-type="equation">δ

.sub.k =(t.sub.k -O.sub.k)(1+O.sub.k)(1-O.sub.k)

and for the inner layer j, ##EQU6## wherein O_i, O_j, and O_k are the outputs of the neural elements in the i, j and k layers of said neural network respectively, t_k is the target value of the output layer neural element of said neural network, h_p is the hidden layer acceleration parameter, W_kj and W_ji are the neural weight values between the k and j layers and between the j and i layers of said neural network respectively, and η

is the learning rate parameter.

3. The method of claim 2 for adaptive color scanning/printing data correction, wherein said learning sample at the time when the learning error reaches below said first predetermined value is extracted with its grey content to constitute a neutral learning sample for feeding back to said neural network for a re-started learning procedure until the learning error reaches again below a second predetermined value.

4. The method of claim 2 for adaptive color scanning/printing data correction, further comprising the step of utilizing said neural network to perform color space transformation.

5. The method of claim 3 for adaptive color scanning/printing data correction, further comprising the step of utilizing said neural network to perform color space transformation.

6. The method of claim 2 for adaptive color scanning/printing data correction, wherein the inner layer acceleration parameter of said back-propagation algorithm (h_p) is in the range of 1.5 to 8, and the learning rate parameter (η

) is in the range of 0.0005 to 0.02.

7. The method of claim 1 for adaptive color scanning/printing data correction, wherein said learning sample at the time when the learning error reaches below a first predetermined value is extracted with its grey content to constitute a neutral learning sample for feeding back to said neural network for a re-started learning procedure until the learning error reaches again below a second predetermined value.

8. The method of claim 7 for adaptive color scanning/printing data correction, further comprising the step of utilizing said neural network to perform color space transformation.

9. The method of claim 1 for adaptive color scanning/printing data correction, further comprising the step of utilizing said neural network to perform color space transformation.

10. The method of claim 1 for adaptive color scanning/printing data correction, wherein said characteristics extracting neural network constrains each picture element neighborhood for learning sample parameters in three-dimensional color space having a range above 0.5 (L_th), 1(a_th), and 1(b_th) and below 5(L_th), 7(a_th), and 7(b_th).

11. The method of claim 1 for adaptive color scanning/printing data correction, wherein said neural network is a multi-layered network comprising 4 to 8 layers and each layer having 20 to 100 neural elements.

12. A method for adaptive color scanning/printing data correction by neural network back-propagation algorithm, wherein a color scanning/printing system includes a color image inputting device, a neural network, and a color image outputting device, said inputting device being connected to said neural network, and said neural network being connected to said outputting device, said method comprising the steps of:

(a) providing pre-selected test color image patterns to said color image outputting device for outputting a first output image;

(b) inputting said first output image to said inputting device to obtain color data of said first output image, and processing a learning sample generated with said color data by said neural network;

(c) said neural network performing a back-propagation balancing algorithm for a learning procedure until the learning error reaches below a predetermined value, wherein said back-propagation algorithm is defined;

for output layer k of said neural network,

space="preserve" listing-type="equation">Δ

W.sub.kj =η

δ

.sub.k.O.sub.j

space="preserve" listing-type="equation">δ

.sub.k =(t.sub.k -O.sub.k)(1+O.sub.k)(1-O.sub.k)

and for the hidden layer j, ##EQU7## wherein O_i, O_j, and O_k are the outputs of the neural elements in the i, j and k layers of said neural network respectively, t_k is the target value of the output layer neural element of said neural network, h_p is the hidden layer acceleration parameter, W_kj and W_ji are the neural weight values between the k and j layers and between the j and i layers of said neural network respectively, and n is the learning rate parameter.

13. The method of claim 12 for adaptive color scanning/printing data correction, wherein said learning sample at the time when the learning error reaches below a first predetermined value is extracted with its grey content to constitute a neutral learning sample for feeding back to said neural network for a re-started learning procedure until the learning error reaches again below a second predetermined value.

14. The method of claim 13 for adaptive color scanning/printing data correction, further comprising the step of utilizing said neural network to perform color space transformation.

15. The method of claim 12 for adaptive color scanning/printing data correction, further comprising the step of utilizing said neural network to perform color space transformation.

16. A method for adaptive color scanning/printing data correction, wherein a color scanning/printing system includes a color image inputting device, a neural network, and a color image outputting device, said inputting device being connected to said neural network, and said neural network being connected to said outputting device, said method comprising the steps of:

providing pre-selected test color image patterns to said color image outputting device for outputting a first output image;

inputting said first output image to said inputting device to obtain color data of said first output image, and processing a learning sample generated with said color data by said neural network;

said neural network processing said constituted learning sample, and at the time when the learning error reaches below a first predetermined value, said learning sample being extracted with its grey content to constitute a neutral learning sample for feeding back to said neural network for a re-started learning procedure until the learning error reaches again below a second predetermined value.

17. A neural network color data correction apparatus for adaptive color scanning/printing, wherein a color scanning/printing system includes a color image inputting device, a neural network, and a color image outputting device, said inputting device being connected to said neural network, and said neural network being connected to said outputting device, said neural network color data correction apparatus comprising:

means for providing pre-selected test color image patterns to said color image outputting device for outputting a first output image;

means for inputting said first output image to said inputting device to obtain color data of said first output image, and processing a learning sample generated with said color data by said neural network means for said neural network to subsequently take the average of the data values of all the color picture elements within a neighborhood in the predetermined vicinity of each input color picture element, and take the average data value of all the color picture elements within the corresponding neighborhood of the corresponding output color picture element, so as to constitute a learning sample pair including the averaged color data values of input and output gamuts respectively; and

means for said neural network to process said learning sample pair to perform a neural network learning procedure.