Fast face beautifying method for digital images

• US 9,501,843 B2
• Filed: 12/18/2014
• Issued: 11/22/2016
• Est. Priority Date: 09/22/2014
• Status: Active Grant

• Current Assignee: Xiamen Meitu Technology Co., Ltd.
• Original Assignee: Xiamen Meitu Technology Co., Ltd.
• Inventors: Zhang, Wei, Fu, Songlin, Zhang, Changding
• Primary Examiner(s): Allison, Andrae S
• Application Number:
  US14/574,815
• Publication Number:
  US 20160086355A1
• Time in Patent Office: 705 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06T 11/001 Texturing; Colouring; Generation of texture or colour inpainting G06T5/77
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06T 5/70 Denoising; Smoothing
• CPC Class Codes: G06V 40/162 using pixel segmentation or colour matching
• CPC Class Codes: H04N 23/83 specially adapted for colour signals
• CPC Class Codes: H04N 23/84 for processing colour signals
• CPC Class Codes: H04N 9/64 Circuits for processing colour signals H04N9/77 takes precedence; camera processing pipelines for processing colour signals H04N23/84
• CPC Class Codes: H04N 9/646 for image enhancement, e.g. vertical detail restoration, cross-colour elimination, contour correction, chrominance trapping filters
• CPC Class Codes: H04N 9/69 for modifying the colour signals by gamma correction controlling camera response for colour signals H04N23/83

Abstract

The present invention provides a fast face beautifying method for digital images. Gaussian blur is performed to an original image to extract a green channel value which is then subject to linear light blending and hard light blending, and a blended green channel value is recalculated. Meanwhile, skin recognition and whitening are performed to the original image, and finally, by using a product of the recalculated green channel value by a probability obtained by skin recognition as a transparency, transparency blending is performed to the original image and the whitened image to compose a beautified image. Moreover, by performing skin recognition to the original image, black pixels will be prevented from being processing by an algorithm, so that hairs, eyes and other non skin parts will be prevented from being processing. Consequently, the final effect of beautification will become better and more natural.

Claims

1. A fast face beautifying method for digital images, comprising the following steps of:

step 1;

reading an original image locally or remotely;

step 2;

performing Gaussian blur to the original image to obtain a blurred image;

step 3;

sequentially extracting a green channel value G of a single pixel of the original image, and performing linear light blending to the green channel with a corresponding pixel of the blurred image to obtain a first green channel value G1;

step 4;

performing continuous hard light blending to the first green channel value G1 obtained by linear light blending with the its own G1 to obtain a second green channel value G2;

step 5;

combining the second green channel value G2 with a red channel value R and a blue channel value B both obtained by Gaussian blur, to obtain a third green channel value G3;

step 6;

performing color mapping to the original image to obtain a whitened image;

step 7;

performing skin color recognition to the original image to obtain a corresponding skin color probability value; and

step 8;

using a product of the third green channel value G3 by the corresponding skin color probability value as a transparency, performing transparency blending to the original image and the whitened image to compose a beautified image.

2. The fast face beautifying method for digital images according to claim 1, wherein the linear light blending in step 3 is performed by the following formula:

G1=(2*G×

21g+1)/2,

wherein, G1 is a color value of a green channel of a single pixel after the linear light blending, G is a color value of a green channel of the original image of the single pixel, and fg is a color value of a green channel of a pixel in the image subjected to Gaussian blur in step 2 corresponding to a same position.

3. The fast face beautifying method for digital images according to claim 1, wherein the continuous hard light blending in step 4 is performed for 1 to 10 times.

4. The fast face beautifying method for digital images according to claim 3, wherein the hard light blending in step 4 is performed by the following formula:

resultColor=((base)<

=128?(base)*(base)/128;

255−

(255−

(base))*(255−

(base))/128),

where, resultColor is a result of the hard light calculation, and (base) is G1 obtained by the linear light blending in step 3.

5. The fast face beautifying method for digital images according to claim 1, wherein the calculation method in step 5 is shown as follows:

6. The fast face beautifying method for digital images according to claim 1, wherein, in step 6, color mapping is performed to the original image to obtain a whitened image, the color mapping is performed by the following formula:

oralColor=arrayCurve[oralColor],

wherein, arrayCurve is a group of predefined color mapping, and oralColor is a color value of a red channel, a green channel and a blue channel of a single pixel in the original image.

7. The fast face beautifying method for digital images according to claim 1, wherein, the performing skin color recognition to the original image to obtain a corresponding skin color probability in step 7 further comprises the following steps of:

step 71;

performing face recognition to the original image to obtain a face region;

step 72;

performing average calculation to the face region to obtain an average skin color;

step 73;

calculating a skin color probability mapping table of the current image according to the average skin color; and

step 74;

performing skin color recognition to the current image according to the skin color probability mapping table to obtain a skin color probability value of the current image.

8. The fast face beautifying method for digital images according to claim 7, wherein step 72 further comprises:

step 721;

initializing an original skin model;

step 722;

calculating an average color value of the whole image as a threshold of the initial skin; and

step 723;

calculating the average skin color of the face region according to the obtained threshold of the initial skin.

9. The fast face beautifying method for digital images according to claim 8, wherein step 722 further comprises:

step 7221;

traversing pixel points of the whole image, and accumulating color values of the red channel, the green channel and the blue channel to obtain an accumulated color sum; and

step 7222, dividing the accumulated color value by the total number of the pixel points to obtain average values of the red channel, the green channel and the blue channel, and using the average values as the threshold of the initial skin.

10. The fast face beautifying method for digital images according to claim 8, wherein step 723 further comprises:

step 7231;

calculating a grayscale value of the average skin color according to the following formula;

GRAY1=0.299*RED+0.587*GREEN+0.114*BLUE,

where, GRAY1 is a gray value of the current pixel point of a gray image, and RED, GREEN and BLUE are color values of red, green and blue channels of the current pixel point of the image, respectively;

step 7232;

using the grayscale value as a threshold for excluding a non-skin portion of the face region;

step 7233;

sequentially traversing the color values of the pixel points within the face region, and obtaining the average skin color according to the following formula;

skin=SkinModel[red][blue],

wherein, skin is a skin value after the color mapping of a skin model, SkinModel is an initialized original skin model, red is the color value of the red channel, and blue is the color value of the blue channel.

11. The fast face beautifying method for digital images according to claim 7, wherein, in step 73, a skin color probability mapping table of the current image is calculated according to the average skin color, where, the skin color probability mapping table is acquired by the following steps of:

step 731;

establishing a skin color probability mapping table 256*256 in size;

step 731;

sequentially performing value assignment to the skin color probability mapping table, the specific pseudo-codes shown as follows;

presetting temporary variables, i.e., i, j, SkinRed_Left, AlphaValue, Offset, TempAlphaValue and OffsetJ, all integers;

presetting a variable of the skin color probability mapping table SkinProbability[256][256]

wherein, SkinRed is the average value of the red channel obtained in step 7222, and SkinBlue is the average value of the blue channel obtained in step 7222;

presetting the value of the SkinRed_Left by the following formula;

kinRed_=SkinRed-128;

For(i=0;

i<

256;

i++)

{;

calculating the value of Offset by the following formula Offset=max(0,min(255, i-SkinRed_Left));

judging whether the value of Offset is less than 128;

if the value of Offset is less than 128, AlphaValue=Offset*2; and

if the value of Offset is greater than or equal to 128, AlphaValue=255;

For(i=0;

j<

256;

j++)

{;

calculating the value of OffsetJ by the following formula OffsetJ=max(0, j−

SkinBlue);

calculating the value of TempAlphaValue by the following formula TempAlphaValue=max(AlphaValue−

(OffsetJ*2),

     0);

judging the value of TempAlphaValue, where, the value of SkinProbability[i][j] is 255 if the value of TempAlphaValue is greater than 160;

the value of SkinProbability[i][j] is 0 if the value of TempAlphaValue is less than 90;

or, the value of SkinProbability[i][j] is equal to TempAlphaValue plus 30;

}

}.

12. The fast face beautifying method for digital images according to claim 7, wherein, in step 74, skin color recognition is performed to the current image according to the skin color probability mapping table to obtain a skin color probability value of the current image, wherein, the calculation method is as follows:

skinColor=SkinProbability[red][blue],

wherein, skinColor is a skin color probability value of the current image, SkinProbability is the skin color probability table, red is the color value of the red channel of the pixel point, and blue is the color value of the blue channel of the pixel point.

13. The fast face beautifying method for digital images according to claim 7, wherein, in step 71, face recognition is performed to the original image to obtain a face region, and the whole image is defined as the face region if the face region recognition is failed.

14. The fast face beautifying method for digital images according to claim 1, wherein, in step 8, the product of multiplying the third green channel value G3 by the corresponding skin color probability value is used as a transparency, and transparency blending is performed to the original image and the whitened image to compose a beautified image, wherein, the formula is as follows:

resultColor=oralColor*alpha+(1.0−

alpha)*arrayColor,

where, resultColor is a color value of the processed beautified image, oralColor is a color value of the original image, arrayColor is a color value of the whitened image obtained in step 6, and alpha is a product of a normalized value of G3 obtained in step 5 by the corresponding skin color probability value, where, the normalization is performed by the following formula;

G3/255.0.