Method of image processing and electronic device utilizing the same

• US 7,345,691 B2
• Filed: 12/02/2004
• Issued: 03/18/2008
• Est. Priority Date: 12/02/2004
• Status: Active Grant

• Current Assignee: Winbond Electronics Corporation
• Original Assignee: Winbond Electronics Corporation
• Inventors: Chien, Chih-Chang
• Primary Examiner(s): Sajous, Wesner
• Application Number:
  US11/000,981
• Publication Number:
  US 20060120596A1
• Time in Patent Office: 1,202 Days
• Field of Search: 345/589, 345/593, 345/597, 345/601, 345/602, 345/612, 345/536, 345/549, 345/555, 382155-157, 382162-167, 700 47- 48, 706 15- 16, 706/33, 358515-519, 358537-538, 348/453, 348/496, 348/577, 348/582, 348/599, 348/612
• US Class Current: 345/589
• CPC Class Codes: G06T 9/002 using neural networks
• CPC Class Codes: H04N 9/643 Hue control means, e.g. flesh tone control

Abstract

A method of image processing using neural networks. Images to be processed and displayed are received. Each image is divided into sections, each comprising pixels represented by color values. A neural network training procedure is executed for the sections to obtain color value tables and mapping tables. Each color value table comprises colors represented by the color values. The mapping tables record the relations between each pixel and the colors in the color value tables. The color value tables and the mapping tables are recorded in an electronic device for image display.

Claims

1. A method of image processing using neural networks, comprising:

receiving at least one image;

dividing each image into at least one section, each comprising pixels represented by at least one color value; and

executing a neural network training procedure for the at least one section to obtain at least one color value table and at least one mapping table, wherein each color value table comprises colors represented by the color values, and the at least one mapping table records the relationships between each pixel and the colors comprised in the at least one color value table, wherein the neural network training procedure comprises;

determining a number of the at least one color value table; and

establishing a neural network containing at least one child neural network, wherein each child neural network corresponds to one color value table and comprises an output layer and an input layer, the output layer contains output nodes as the number of colors contained in the corresponding color value table, the input layer contains input node sets as the number of the pixels contained in each section, and each input node set has input nodes as the number of the color values.

2. The method of image processing using neural networks as claimed in claim 1, further comprising:

storing the at least one color value table and the at least one mapping table in an electronic device; and

the electronic device displaying the images according to the at least one color value table and the at least one mapping table.

3. The method of image processing using neural networks as claimed in claim 1, wherein the neural network training procedure comprises:

inputting the color values of the pixels contained in each section to the input nodes;

adjusting the color values of the pixels to obtain at least one output value;

calculating an error value of the pixels for each section according to the color values and the output values, and outputting the error values by the output nodes;

recording the output values in the at least one color value table if the error values satisfy a predetermined error condition; and

repeating the adjustment and the calculation if the error values do not satisfy the predetermined error condition.

4. The method of image processing using neural networks as claimed in claim 3, wherein error values are calculated by:

$E=\sqrt{\frac{{\left(R-r_t\right)}^2+{\left(G-g_t\right)}^2+{\left(B-b_t\right)}^2}{3}}$

wherein E is the error value, R, G, B are the color values of the pixels, r_t, g_t, b_t are the output values.

5. The method of image processing using neural networks as claimed in claim 4, wherein adjustment is accomplished by:

$\mathrm{\Delta <br><br>}<br><br><br><br>r=\frac{\partial <br><br>E}{\partial <br><br>r}=-\frac{R-r_t}{3\times <br><br>E},r_{t+1}=r_t+\mathrm{\mu <br><br>}\times <br><br>\mathrm{\Delta <br><br>}<br><br><br><br>r$ $\mathrm{\Delta <br><br>}<br><br><br><br>g=\frac{\partial <br><br>E}{\partial <br><br>g}=-\frac{G-g_t}{3\times <br><br>E},g_{t+1}=g_t+\mathrm{\mu <br><br>}\times <br><br>\mathrm{\Delta <br><br>}<br><br><br><br>g$ $\mathrm{\Delta <br><br>}<br><br><br><br>b=\frac{\partial <br><br>E}{\partial <br><br>b}=-\frac{B-b_t}{3\times <br><br>E},b_{t+1}=b_t+\mathrm{\mu <br><br>}\times <br><br>\mathrm{\Delta <br><br>}<br><br><br><br>b$

wherein E is the error formula, R, G, B are the color values of the pixels, r_t, g_t, b _t are the output values, t is calculation count, μ

is an adjustment rate parameter.

6. The method of image processing using neural networks as claimed in claim 3, wherein in the input step the color value of the pixels are further normalized 0 to 1.

7. The method of image processing using neural networks as claimed in claim 3, wherein in the input step the output values are further randonmized for initialization.

8. An electronic device with image processing function using a neutral network, comprising:

a storage unit for storing at least one color value table and at least one mapping table, wherein the at least one color value table, and the at least one mapping table are obtained by a neural network training procedure, wherein the at least one mapping table records the relationships between each pixel and the colors comprised in the at least one color value table, wherein the neural network procedure further comprise;

determining number of the at least one color value table; and

establishing a neural network containing at least one child neural network, wherein each child neural network corresponds to one color value table and comprises an output layer and an input layer, the output layer contains output nodes as the number of colors contained in the corresponding color value table, the input layer contains input node sets as the number of the pixels contained in each section, and each input node set has input nodes as the number of the color values; and

a processing unit, coupled to the storage unit, for displaying at least one image according to the least one color value table and the at least one mapping table.

9. The electronic device with image processing using a neutral network as claimed in claim 8, wherein the neural network training procedure is executed by a training module, and comprises:

receiving at least one image;

dividing each image into at least one section, wherein each section comprises pixels represented by at least one color value; and

obtaining at least one color value table and at least one mapping table, wherein each color value table comprises colors represented by the color values, and the at least one mapping table records the relations between each pixel and the colors comprised in the at least one color value table.

10. The electronic device with image processing using a neutral network as claimed in claim 8, wherein the neural network procedure further comprises:

inputting the color values of the pixels contained in each section to the input nodes;

adjusting the color values of the pixels to obtain at least one output value;

calculating an error value of the pixels for each section according to the color values and the output values, and outputting the error values by the output nodes;

recording the output values in the at least one color value table if the error values satisfy a predetermined error condition; and

repeating the adjustment and the calculation if the error values do not satisfy the predetermined error condition.

11. The electronic device with image processing using a neutral network as claimed in claim 10, wherein the error values are calculated by:

$E=\sqrt{\frac{{\left(R-r_t\right)}^2+{\left(G-g_t\right)}^2+{\left(B-b_t\right)}^2}{3}}$

wherein E is the error value, R, G, B are the color values of the pixels, r_t, g_t, b_t are the output values.

12. The electronic device with image processing using a neutral network as claimed in claim 11, wherein the adjustment is accomplished by:

$\mathrm{\Delta <br><br>}<br><br><br><br>r=\frac{\partial <br><br>E}{\partial <br><br>r}=-\frac{R-r_t}{3\times <br><br>E},r_{t+1}=r_t+\mathrm{\mu <br><br>}\times <br><br>\mathrm{\Delta <br><br>}<br><br><br><br>r$ $\mathrm{\Delta <br><br>}<br><br><br><br>g=\frac{\partial <br><br>E}{\partial <br><br>g}=-\frac{G-g_t}{3\times <br><br>E},g_{t+1}=g_t+\mathrm{\mu <br><br>}\times <br><br>\mathrm{\Delta <br><br>}<br><br><br><br>g$ $\mathrm{\Delta <br><br>}<br><br><br><br>b=\frac{\partial <br><br>E}{\partial <br><br>b}=-\frac{B-b_t}{3\times <br><br>E},b_{t+1}=b_t+\mathrm{\mu <br><br>}\times <br><br>\mathrm{\Delta <br><br>}<br><br><br><br>b$

wherein E is the error formula, R, G, B are the color values of the pixels, r_t, g_t, b_t are the output values, t is calculation counts, μ

is a adjustment rate parameter.

13. The electronic device with image processing using a neutral network as claimed in claim 10, wherein the training module further normalizes the color values of the pixels from 0 to 1.

14. The electronic device with image processing using a neutral network as claimed in claim 10, wherein the training module further randomizes the output values for initialization.

15. The electronic device with image processing using a neutral network as claimed in claim 9, wherein the training module is a computer.

16. The electronic device with image processing using a neutral network as claimed in claim 9, wherein the storage unit is a chip.

17. The electronic device with image processing using a neutral network as claimed in claim 9, wherein the processing unit is a central processing unit or a controller.