CN100571404C - Skin color signal correcting method - Google Patents

Abstract

Skin color signal correcting method relates to image processing method, particularly a kind of skin color signal correcting method.The present invention is directed in the flesh correction luminance signal to the influence of the colour of skin, propose skin color signal correcting method.Technical scheme of the present invention may further comprise the steps: luminance signal and the color difference signal of importing image to be corrected; Value according to luminance signal is done preliminary treatment to color difference signal; Judge whether pretreated color difference signal is in the colour of skin distributed areas, if then pretreated color difference signal is made the color difference signal after the output luminance signal and flesh correction behind the flesh correction; Otherwise directly export luminance signal and the color difference signal of being imported.The invention has the beneficial effects as follows, solved effectively and made the problem that patch occurs in the flesh correction processing, set different correction coefficient, realized adaptive real-time flesh correction according to the detected colour of skin depth to crossing bright or dark excessively skin.And whole calculating process is simple, is convenient to hardware and realizes.

Description

Skin color signal correcting method

Technical field

The present invention relates to image processing method, particularly a kind of skin color signal correcting method.

Background technology

ITU-R BT.601 is that (standarddefinition television, SDTV) image digitazation standard are used for the television image of 525 scan lines and 625 scan lines is carried out digital coding an international standard definition television.The YCbCr color space is the part of ITU-RBT-601, wherein Y is a luminance signal, be defined as [16,235] 8 bit binary data in the scope, Cb is a color difference signal, is defined as [16,240] 8 bit binary data in the scope, Cr is a color difference signal, is defined as 8 bit binary data in [16,240] scope.

Flesh correction is just becoming a critical function module in the video image process chip, and it can detect the colour of skin (exposed skin) of people in the image in real time, and proofreaies and correct, and makes it the truer colour of skin near the people.In the YCbCr color space, colour of skin picture element always accumulates in certain spatial dimension, i.e. colour of skin distributed areas, and as shown in Figure 1 in the accompanying drawing, therefrom as can be seen, along with the difference of Y, the distribution of Cb, Cr is discrepant.And present a lot of algorithms all only are conceived to color difference signal itself is handled, do not consider the influence of luminance signal, this just causes crossing bright or dark excessively skin treatments effect undesirable, occurs patch etc. easily, can not proofread and correct adaptively according to the depth of the colour of skin.And, just detect area of skin color usually based on the edge detection algorithm in HSI space, do not do flesh correction, do not consider the influence of luminance signal to the colour of skin yet, and relate to complex mathematical computational process, so the hardware realization is very difficult, is not easy to be used in the image real time processing system.

Summary of the invention

Technical problem to be solved by this invention is at the influence of luminance signal in the flesh correction to the colour of skin, to propose skin color signal correcting method.

The technical scheme that the present invention solve the technical problem employing is that skin color signal correcting method may further comprise the steps: (1) input image brightness signal to be corrected and color difference signal; (2) according to the value of luminance signal color difference signal is done preliminary treatment:

1. when 16≤Y≤120:

$\mathrm{Cb}\_\mathrm{deal}=m1+\frac{47\×(\mathrm{Cb}-m1)}{j1}$

$\mathrm{Cr}\_\mathrm{deal}=n1+\frac{39\×(\mathrm{Cr}-n1)}{k1}$

$m1=108+\frac{10\×(120-Y)}{128}$ $j1=23+\frac{24\×(Y-16)}{128}$

In the formula $n1=154-\frac{10\×(120-Y)}{128}$ $k1=20+\frac{19\×(Y-16)}{128}$

2. when 120＜Y＜180:

Cb_deal＝Cb

Cr_deal＝Cr

3. when Y 〉=180:

$\mathrm{Cb}\_\mathrm{deal}=m2+\frac{47\×(\mathrm{Cb}-m2)}{j2}$

$\mathrm{Cr}\_\mathrm{deal}=n2+\frac{39\×(\mathrm{Cr}-n2)}{k2}$

In the formula $m2=108+\frac{10\×(Y-180)}{64}$ $j2=14+\frac{32\×(235-Y)}{64}$

$n2=154+\frac{22\×(Y-180)}{64}$ $k2=10+\frac{32\×(235-Y)}{64}$

In the following formula, Y is a luminance signal, and Cb, Cr are color difference signal, and Cb_deal, Cr_deal are pretreated color difference signal;

(3) judge whether pretreated color difference signal is in the area of skin color, if then enter step (4), otherwise enter step (6): Rule of judgment is: when Cb_deal, Cr_deal satisfy $\frac{3\&times;{(x-1.6)}^2}{2048}+\frac{5\&times;{(y-2.4)}^2}{1024}<1$ The time, pretreated color difference signal is in the colour of skin distributed areas, otherwise just is not in the colour of skin distributed areas,

x＝-0.819×(Cb_deal-109)+0.574×(Cr_deal-152)

In the formula,

y＝-0.574×(Cb_deal-109)-0.819×(Cr_deal-152)

In the following formula, Cb_deal, Cr_deal are pretreated color difference signal;

(4) pretreated color difference signal is made flesh correction:

$\mathrm{Cb}\_\mathrm{out}=\mathrm{Cb}\_\mathrm{deal}+\frac{(\mathrm{Cb}\_\mathrm{deal}-\mathrm{mid}\_\mathrm{cb})}{\mathrm{ftc}\_\mathrm{coefficient}}$

$\mathrm{Cr}\_\mathrm{out}=\mathrm{Cr}\_\mathrm{deal}+\frac{(\mathrm{Cr}\_\mathrm{deal}-\mathrm{mid}\_\mathrm{cr})}{\mathrm{ftc}\_\mathrm{coefficient}}$

In the following formula, Cb_deal, Cr_deal is pretreated color difference signal, Cb_out is the value behind the Cb_deal flesh correction, Cr_out is the value behind the Cr_deal flesh correction, mid_cb is the pairing Cb value of the central value of colour of skin distributed areas, mid_cr is the pairing Cr value of the central value of colour of skin distributed areas, draw by a large amount of experiments, Cb in the colour of skin distributed areas, the Cr span is generally Cb ∈ [75,128], Cr ∈ [130,165], central value (the mid_cb of colour of skin distributed areas, mid_cr) Dui Ying Cb, the Cr span is respectively: mid_cb ∈ [100,110], mid_cr ∈ [130,140], ftc_coefficient is a colour of skin reinforcing coefficient, and value is 2 or 4;

(5) output luminance signal, the color difference signal behind the flesh correction;

(6) luminance signal and the color difference signal of output step (1) input.

By getting different ftc_coefficient values, the further strength grade of refinement flesh correction.To being in different spans of Cb_deal, Cr_deal in the area of skin color as Rule of judgment to the concrete value of ftc_coefficient, as: in Cb_deal 〉=105 and Cr_deal 〉=140 o'clock, ftc_coefficient=2, and for Cb_deal, the Cr_deal of other scope, then get ftc_coefficient=4, thus the colour of skin adaptively in the correcting image.By actual verification, mid_cb=105, mid_cr=130, during ftc_coefficient=4, the result of image processing is more satisfactory.

The invention has the beneficial effects as follows, considered the influence of luminance signal to the colour of skin, solved effectively and made the problem that patch occurs in the flesh correction processing crossing bright or dark excessively skin, set different correction coefficient according to the detected colour of skin depth, realized adaptive real-time flesh correction, and whole calculating process is simple, is convenient to hardware and realizes, can be widely used in the image quality improving chip of video image processing.

Description of drawings

Fig. 1 is the three-dimensional cluster feature figure of the colour of skin in the YCbCr space.

Fig. 2 is the flow chart of algorithm of the present invention.

Embodiment

This algorithm is optimized and verified that the image pixel of selection is 1366*768 by Matlab6.5 programming.At first, adopt the imread () function of Matlab6.5 to read in image, and process rgb2ycbcr () function arrives the YCbCr color space with image transitions, isolate Y, Cb, three signal components of Cr, carry out flesh correction with foregoing algorithmic formula again, pixel value after will proofreading and correct by ycbcr2rgb () function is at last changed back rgb color space, and shows image after flesh correction is handled with imshow () function.

Provide specific embodiments of the invention below in conjunction with concrete input data.

Specific embodiment 1: input Y=80, Cb=101, Cr=145, because 16≤Y≤120, so Cb, Cr are carried out preliminary treatment, getting behind the Y=80 substitution preliminary treatment formula:

$m1=108+\frac{10\&times;(120-80)}{128}$ $j1=23+\frac{24\&times;(80-16)}{128}$

$n1=154-\frac{10\&times;(120-80)}{128}$ $k1=20+\frac{19\&times;(80-16)}{128}$

$\mathrm{Cb}\_\mathrm{deal}=m1+\frac{47\&times;(101-m1)}{j1}=98$

$\mathrm{Cr}\_\mathrm{deal}=n1+\frac{39\&times;(145-n2)}{k1}=143$

Whether the value of then judging Cb_deal, Cr_deal is in the colour of skin distributed areas, respectively

x＝-0.819×(98-109)+0.574×(143-152)

y＝-0.574×(98-109)-0.819×(143-152)

The substitution equation can get: $\frac{3\&times;{(x-1.6)}^2}{2048}+\frac{5\&times;{(y-2.4)}^2}{1024}=0.6516<1,$ Then Cb_deal, Cr_deal are in the colour of skin distributed areas, Cb_deal, Cr_deal are done flesh correction, because this moment Cb_deal=98≤105, then get ftc_coefficient=4, Cb_deal=98, Cr_deal=143, mid_cb=105, mid_cr=130, ftc_coefficient=4 substitution flesh correction formula can get:

$\mathrm{Cb}\_\mathrm{out}=98+\frac{(98-105)}{4}=100$

$\mathrm{Cr}\_\mathrm{out}=143+\frac{(143-130)}{4}=149$

Export 80,100,149 then, compare with Y=80, Cb=101, the Cr=145 of input,

Make after handling like this to be in dark skin pixel value and to have obtained correction, the image that show this moment is then more near people's the true colour of skin.

Specific embodiment 2: input Y=223, Cb=106, Cr=136, because Y 〉=180, so Cb, Cr are carried out preliminary treatment, getting behind the Y=223 substitution preliminary treatment formula:

$m2=108+\frac{10\&times;(223-180)}{64}$ $j2=14+\frac{32\&times;(235-223)}{64}$

$n2=154+\frac{22\&times;(223-180)}{64}$ $k2=10+\frac{32\&times;(235-223)}{64}$

$\mathrm{Cb}\_\mathrm{deal}=m2+\frac{47\&times;(106-m2)}{j2}=105$

$\mathrm{Cr}\_\mathrm{deal}=n2+\frac{39\&times;(136-n2)}{k2}=142$

Judge then whether Cb_deal, Cr_deal are in the colour of skin distributed areas, respectively

x＝-0.819×(105-109)+0.574×(142-152)

y＝-0.574×(105-109)-0.819×(142-152)

The substitution equation can get: $\frac{3\&times;{(x-1.6)}^2}{2048}+\frac{5\&times;{(y-2.4)}^2}{1024}=0.3576<1,$ Then Cb_deal, Cr_deal are in the colour of skin distributed areas, Cb_deal, Cr_deal are done flesh correction, because ftc_coefficient=2 item is got in Cb=105 〉=105 and Cr=142 〉=140, Cb_deal=105, Cr_deal=142, mid_cb=105, mid_cr=130, ftc_coefficient=2 substitution flesh correction formula can get:

$\mathrm{Cb}\_\mathrm{out}=105+\frac{(105-105)}{2}=107$

$\mathrm{Cr}\_\mathrm{out}=142+\frac{(142-130)}{2}=139$

Export 223,107,139 then, compare with Y=223, Cb=106, the Cr=136 of input, make after handling like this to be in bright skin pixel value and to have obtained correction, the image of demonstration this moment is then more near people's the true colour of skin.

Specific embodiment 3: input Y=23, Cb=89, Cr=165, because 16≤Y≤120, so Cb, Cr are carried out preliminary treatment, getting behind the Y=23 substitution preliminary treatment formula:

$m1=108+\frac{10\&times;(120-23)}{128}$ $j1=23+\frac{24\&times;(80-23)}{128}$

$n1=154-\frac{10\&times;(120-23)}{128}$ $k1=20+\frac{19\&times;(80-23)}{128}$

$\mathrm{Cb}\_\mathrm{deal}=m1+\frac{47\&times;(89-m1)}{j1}=64$

$\mathrm{Cr}\_\mathrm{deal}=n1+\frac{39\&times;(165-n2)}{k1}=181$

Judge then whether Cb_deal, Cr_deal are in the colour of skin distributed areas, respectively

x＝-0.819×(64-109)+0.574×(181-152)

y＝-0.574×(64-109)-0.819×(181-152)

The substitution equation can get: $\frac{3\&times;{(x-1.6)}^2}{2048}+\frac{5\&times;{(y-2.4)}^2}{1024}=3.9096>1,$ Then Cb_deal, Cr_deal are not in the colour of skin distributed areas, Cb_deal, Cr_deal are not done flesh correction, output 23,89,165 is compared no change with Y=23, Cb=89, the Cr=165 of input, because Cb_deal, Cr_deal do not belong in the colour of skin distributed areas.At this moment, the image no change that shows this moment.

Claims (2)

1, skin color signal correcting method is characterized in that, may further comprise the steps:

(1) input image brightness signal to be corrected and color difference signal;

(2) according to the value of luminance signal color difference signal is done preliminary treatment:

1. when 16≤Y≤120:

$\mathrm{Cb}\_\mathrm{deal}=m1+\frac{47\&times;(\mathrm{Cb}-m1)}{j1}$

$\mathrm{Cr}\_\mathrm{deal}=n1+\frac{39\&times;(\mathrm{Cr}-n1)}{k1}$

$m1=108+\frac{10\&times;(120-Y)}{128}$ $j1=23+\frac{24\&times;(Y-16)}{128}$

In the formula $n1=154-\frac{10\&times;(120-Y)}{128}$ $k1=20+\frac{19\&times;(Y-16)}{128}$

2. when 120＜Y＜180:

Cb_deal＝Cb

Cr_deal＝Cr

3. when Y 〉=180:

$\mathrm{Cb}\_\mathrm{deal}=m2+\frac{47\&times;(\mathrm{Cb}-m2)}{j2}$

$\mathrm{Cr}\_\mathrm{deal}=n2+\frac{39\&times;(\mathrm{Cr}-n2)}{k2}$

In the formula $m2=108+\frac{10\&times;(Y-180)}{64}$ $j2=14+\frac{32\&times;(235-Y)}{64}$

$n2=154+\frac{22\&times;(Y-180)}{64}$ $k2=10+\frac{32\&times;(235-Y)}{64}$

In the following formula, Y is a luminance signal, and Cb, Cr are color difference signal, and Cb_deal, Cr_deal are pretreated color difference signal;

(3) judge whether pretreated color difference signal is in the area of skin color, if then enter step (4), otherwise enter step (6): Rule of judgment is: when Cb_deal, Cr_deal satisfy $\frac{3\&times;{(x-1.6)}^2}{2048}+\frac{5\&times;{(y-2.4)}^2}{1024}<1$ The time, pretreated color difference signal is in the colour of skin distributed areas, otherwise just is not in the colour of skin distributed areas,

x＝-0.819×(Cb_deal-109)+0.574×(Cr_deal-152)

In the formula,

y＝-0.574×(Cb_deal-109)-0.819×(Cr_deal-152)

In the following formula, Cb_deal, Cr_deal are pretreated color difference signal;

(4) pretreated color difference signal is made flesh correction:

$\mathrm{Cb}\_\mathrm{out}=\mathrm{Cb}\_\mathrm{deal}+\frac{(\mathrm{Cb}\_\mathrm{deal}-\mathrm{mid}\_\mathrm{cb})}{\mathrm{ftc}\_\mathrm{coeffi\; cient}}$

$\mathrm{Cr}\_\mathrm{out}=\mathrm{Cr}\_\mathrm{deal}+\frac{(\mathrm{Cr}\_\mathrm{deal}-\mathrm{mid}\_\mathrm{cr})}{\mathrm{ftc}\_\mathrm{coeffi\; cient}}$

In the following formula, Cb_deal, Cr_deal are pretreated color difference signal, Cb_out is the value behind the Cb_deal flesh correction, and Cr_out is the value behind the Cr_deal flesh correction, and mid_cb is the pairing Cb value of the central value of colour of skin distributed areas, mid_cr is the pairing Cr value of the central value of colour of skin distributed areas, described mid_cb, mid_cr scope are respectively: mid_cb ∈ [100,110], mid_cr ∈ [130,140], ftc_coefficient is a colour of skin reinforcing coefficient, and value is 2 or 4;

(5) output luminance signal, the color difference signal behind the flesh correction;

(6) luminance signal and the color difference signal of output step (1) input.

2, skin color signal correcting method as claimed in claim 1 is characterized in that, described mid_cb is 105, and mid_cr is 130.

Images