CN103186785B - Skin color detection method and system - Google Patents
Skin color detection method and system Download PDFInfo
- Publication number
- CN103186785B CN103186785B CN201210331409.2A CN201210331409A CN103186785B CN 103186785 B CN103186785 B CN 103186785B CN 201210331409 A CN201210331409 A CN 201210331409A CN 103186785 B CN103186785 B CN 103186785B
- Authority
- CN
- China
- Prior art keywords
- skin
- colour
- correction
- image
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title abstract description 17
- 238000012937 correction Methods 0.000 claims abstract description 153
- 238000000034 method Methods 0.000 claims description 61
- 238000012545 processing Methods 0.000 claims description 41
- 238000002372 labelling Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 9
- 238000010801 machine learning Methods 0.000 claims description 8
- 238000013139 quantization Methods 0.000 claims description 8
- 230000005540 biological transmission Effects 0.000 claims description 7
- 235000013399 edible fruits Nutrition 0.000 claims 2
- 230000007246 mechanism Effects 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000012217 deletion Methods 0.000 description 3
- 230000037430 deletion Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 239000002023 wood Substances 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 239000012535 impurity Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035807 sensation Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000000392 somatic effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Image Processing (AREA)
- Processing Of Color Television Signals (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a skin color detection method and a system, wherein the skin color detection method comprises the following steps: obtaining an image; performing parameter correction on the image according to the correction reference; according to the pixels of the image, performing table lookup on the skin color probability statistical table to generate a table lookup result; judging the table look-up result according to at least one automatic correction criterion to generate a judgment result; and correcting the skin color probability statistical table according to the judgment result.
Description
Technical field
The invention relates to a kind of method and device, and in particular to a kind of colour of skin method for detecting and skin
Color detector.
Background technology
Colour of skin detecting belongs to a kind of image processing category, and the pixel in the either statically or dynamically picture of can be used for classifying is
No for the colour of skin.Owing to Skin Color Information can be used to judge face, hands, body position, so applicable scope
Quite extensive, such as: the identification of human face detection, gesture, finger tip detecting, porny identification, somatic sensation television game,
Human-computer interaction etc..
It is well known, however, that technology be when off-line use statistical to produce colour of skin probability statistical table, but
When using aforementioned colour of skin probability statistical table as the foundation of skin color classification, its detecting result has detecting erroneous judgement
(false positive) and detecting leakage (miss detection), cause noise that the colour of skin detects and imperfect.
Summary of the invention
One purpose of present invention is to provide a kind of colour of skin method for detecting and colour of skin detecting system, its school
Reduction colour of skin detecting False Rate and the detecting leak rate that just module can be instant.Furthermore can be according to detecting result
Feedback correction picture, and the manually or automatically newly-increased or deletion colour of skin, expand or shrink the model of the detecting colour of skin
Enclose.
For reaching above-mentioned purpose, a technical scheme of present invention is about a kind of colour of skin method for detecting.The colour of skin
Method for detecting comprises the steps of acquirement one image;According to correction reference so that image is carried out parameter correction;
According to the pixel of image, carry out tabling look-up to produce checking result to colour of skin probability statistical table;According at least one certainly
Checking result is judged to produce judged result by dynamic(al) correction criterion;And according to judged result to revise skin
Color probability statistical table.
According to one embodiment of the invention, colour of skin method for detecting also comprises the steps of feedback checking result, and
According to checking result correction correction reference.
According to another embodiment of the present invention, colour of skin method for detecting also comprises the steps of and carries out plurality of pictures
Sampling;The color space of aforementioned plurality of pictures is quantified;And add up each in aforementioned plurality of pictures
The probability that color is the colour of skin, to produce colour of skin probability statistical table.
According to another embodiment of the present invention, colour of skin method for detecting also comprises plurality of pictures carries out manual colour of skin mark
Note, quantifies this manual colour of skin labelling result, and by carrying out add up or in terms of supervised machine learning comes
Calculate the probability that color is the colour of skin of each in aforementioned plurality of pictures, to produce colour of skin probability statistical table.
According to another embodiment of the present invention, colour of skin method for detecting also comprise image is carried out image processing with produce
Image processing result, and according to image processing result to increase newly or to revise from dynamic(al) correction criterion.
According to yet another embodiment of the invention, according to judged result to revise the step of colour of skin probability statistical table, it is
Perform once every a stabilized image number.
According to further embodiment of this invention, according to the colour of skin and non-colour of skin sample, the manual school of user input
Positive colour of skin probability statistical table.
For reaching above-mentioned purpose, another technical scheme of present invention is about a kind of colour of skin detecting system.Skin
Color detecting system comprises image input module, parameter correction module, form access module, checking result storage
Module and automatically correction module.In operation, image input module is in order to obtain image;Parameter correction mould
Block comprises correction reference, and above-mentioned parameter correction module is so that image is carried out parameter correction according to correction reference;
Automatically correction module is in order to basis at least from dynamic(al) correction criterion, judges checking result to produce judgement
As a result, and will determine that form access module is given in result transmission, above table access module according to judged result,
With correct skin tones probability statistical table, and then make judged result next time the most correct.
According to one embodiment of the invention, colour of skin detecting system also comprises feedback correction module, form access module
In order to the pixel according to image, carry out tabling look-up to produce checking result, above-mentioned feedback to colour of skin probability statistical table
Correction module is obtained checking result by form access module, and feedback checking result gives parameter correction module, ginseng
Number correction module can be according to the correction reference of checking result corrected parameter correction module.
According to another embodiment of the present invention, colour of skin detecting system also comprises colour of skin image sample process module, amount
Change module and statistical module.In operation, colour of skin image sample process module is in order to carry out plurality of pictures
Automatic or manual colour of skin labelling;Quantization modules, in order to make to quantify to the color of color space, selects some
Representative color;And statistical module use statistics or the mode of supervised machine learning calculate aforementioned multiple
In picture, each color that represents is marked as the number of times of the colour of skin and the non-colour of skin to calculate its probability as the colour of skin, with
Produce colour of skin probability statistical table.
According to another embodiment of the present invention, colour of skin detecting system also comprises image processing module, in order to image
Carry out image processing to produce image processing result, and provide image processing result to give automatic correction module with newly
Increase or revise from dynamic(al) correction criterion.In operation on, image processing module in order to image is carried out image processing with
Produce image processing result, such as this image processing module can perform the detecting of mobile pixel, human face detection or its
His colour of skin detection techniques etc., and provide the detecting unit that image processing result gives automatic correction module to calculate, meter
Calculation result is stored in data base and detects result for changing, occur simultaneously, join the collection colour of skin, and then through from dynamic(al) correction
Module correction colour of skin probability statistical table, to expand or shrink colour of skin reconnaissance range.
According to yet another embodiment of the invention, the automatic correction module in colour of skin detecting system is every a stabilized image
Number will determine that form access module is given in result transmission, to revise colour of skin probability statistical table.
According to yet another embodiment of the invention, the manual correction module in colour of skin detecting system, can be according to use
The colour of skin or non-colour of skin image that person is provided increase or delete colour of skin color newly so that colour of skin probability statistical table institute
The skin pixel of record can dynamically adjust.
Therefore, according to the technology contents of the present invention, the embodiment of the present invention provide a kind of colour of skin method for detecting and
Colour of skin detecting system, can be by automatic correction module, manual correction module, when colour of skin detecting system and side
When method finds colour of skin detecting mistake, can correct skin tones probability statistical table immediately, make colour of skin probability statistical table more
Completely, so that the judgement of the colour of skin detecting system of the embodiment of the present invention and method is the most accurate, and can lead to
Crossing feedback correction module makes the parameter adjustment inputting image the most correct, in order to allow colour of skin detecting system result
More accurate.
Additionally, due to the colour of skin detecting system of the embodiment of the present invention and method are able to every stabilized image number
Just being revised colour of skin probability statistical table by correction mechanism, therefore, the colour of skin of the embodiment of the present invention is detectd
Examining system and method can revise colour of skin probability statistical table under the situation of not excessive use system resource.
Accompanying drawing explanation
For the above and other purpose of the present invention, feature, advantage can be become apparent with embodiment, appended
Being described as follows of accompanying drawing:
Fig. 1 is the block schematic diagram illustrating a kind of colour of skin detecting system according to one embodiment of the invention;
Fig. 2 is the schematic flow sheet illustrating a kind of colour of skin method for detecting according to another embodiment of the present invention;
Fig. 3 is to illustrate to add up according to colour of skin probability in a kind of colour of skin method for detecting of yet another embodiment of the invention
The production method schematic diagram of table.
[main element symbol description]
200: colour of skin method for detecting
210~280: step
300: the production method of colour of skin probability statistical table
310~330: step
1000: colour of skin detecting system
1010: image input module
1030: parameter correction module
1040: feedback module
1050: Query Result stores module
1060: form access module
1070: colour of skin probability statistical table
1080: correction module automatically
1082: detecting unit
1084: data base
1090: image processing module
1100: manual correction module
1110: colour of skin image sample process module
1120: quantization modules
1130: statistical module
Detailed description of the invention
In order to the narration making the present invention is more detailed with complete, can refer to appended accompanying drawing and the following stated is various
Embodiment, number identical in accompanying drawing represents same or analogous element.But the embodiment provided not is used
To limit the scope that the present invention is contained, and the description of structure operation is not used to limit its order performed, and appoints
The structure what is reconfigured by element, is produced the device with impartial effect, is all what the present invention was contained
Scope.
Wherein accompanying drawing is the most for the purpose of description, and maps not according to life size.On the other hand, it is thus well known that
Element and step are not described in embodiment, to avoid causing the present invention unnecessary restriction.
Fig. 1 is the block schematic diagram illustrating a kind of colour of skin detecting system 1000 according to one embodiment of the invention.
As it can be seen, colour of skin detecting system 1000 comprise image input module 1010, parameter correction module 1030,
Form access module 1060, checking result generation module 1050, feedback correction module 1040, from dynamic(al) correction
Module 1080 and manual correction module 1100.
In operation, image input module 1010 in order to obtain image, so-called image input module 1010
Refer to any equipment being obtained image, such as camera, network disk, database, the right present invention
It is not limited thereto.Parameter correction module 1030 comprises correction reference, above-mentioned parameter correction module 1030
It is so that image is carried out parameter correction according to correction reference.For example, if the parameter set by correction reference
One be brightness value be 130, when the brightness value of image is 50 when, parameter correction module 1030 can root
According to correction reference, the brightness value of image being increased by 80, and make the brightness value of image reach 130, the right present invention is also
Be not limited, parameter correction module 1030 in addition to the brightness value of image, also remaining parameter of recoverable.
Additionally, form access module 1060 is in order to the pixel according to image, to colour of skin probability statistical table 1070
Carrying out tabling look-up to produce checking result, this checking result can be temporarily stored in checking result and store module 1050, or
Person is directly delivered and gives feedback correction module 1040 and automatic correction module 1080, in the present embodiment, table
Lattice access module 1060 can comprise enquiry form, store form, loading form, acquirement color colour of skin machine
Whether rate, setpoint color colour of skin probability, setpoint color are the functions such as the colour of skin, are not limited during actual enforcement.
Automatically correction module 1080 is in order to basis at least from dynamic(al) correction criterion, judges checking result
To produce judged result, and will determine that form access module 1060, above table access mould are given in result transmission
Block 1060 is according to judged result, to revise colour of skin probability statistical table 1070.For example, aforementioned automatic school
Positive criterion can be isolated point (isolated point) determining method, and automatic correction module 1080 judges according to isolated point
Described in as described below section of the mode that checking result is judged by method.
In the picture element matrix of nine grids in aforementioned image, if the checking result of the isolated pixel of centre
It is the colour of skin, and the most non-colour of skin of checking result of eight pixels around, now according to the rule of isolated point determining method
Then, can judge the most non-colour of skin of isolated pixel of centre, above-mentioned judged result is the isolated pixel of centre
Rgb value does not represent the colour of skin, and this judged result can be passed back to form access module 1060, above table
Access module 1060 is according to judged result, by above-mentioned isolated pixel RGB in colour of skin probability statistical table 1070
Value is modified to the non-colour of skin.The most above-mentioned it is not limited to isolated point determining method from dynamic(al) correction criterion, is familiar with this skill
Person when be optionally used suitable judging rules using as above-mentioned from dynamic(al) correction criterion.
Above-mentioned automatic correcting mode can be realized by a series of detecting unit 1082, and it is in order to meter automatically
Calculate false skin pixel (false positive) and false non-skin pixel (miss detection), access through form
Module 1060 correct skin tones probability statistical table.It is exemplified below the sense mode of several detecting unit 1082:
1., in detecting result, a pixel is skin pixel, and its adjacent n-pixel (nine grids) is neither skin
Color pixel, represents that this isolated point (isolated point) is false skin pixel (false positive), then revises this face
Color is the non-colour of skin, and n can be a constant value (such as: 7 or 8).
2., in detecting result, a pixel is non-skin pixel, and its adjacent n-pixel is all skin pixel,
Represent that this isolated point isolated point is false non-skin pixel (miss detection), then can be aided with other again
Rule tests whether this color is the colour of skin, is the colour of skin by then revising this color, and n can be a constant value (example
As: 7 or 8).
3. in detecting result, use noise detection techniques detecting skin pixel, then noise is false colour of skin picture
Element, revising this color is the non-colour of skin.Such as may be used to colour of skin image is carried out remove impurity news, and will go remove impurity
The image of news is deducted image originally and is obtained noise, and colour of skin probability statistical table 1070 can be according to aforementioned shadow
As result is revised.
4., in detecting result, use skin pixel erosion, Extension Operations, then test this with other rules
After expansion, whether color is the colour of skin, is the colour of skin by then revising this color.
Additionally, colour of skin detecting system 1000 also comprises manual correction module 1100, it is in order to according to user
The colour of skin provided or non-colour of skin image increase newly or delete in colour of skin probability statistical table 1070 and be labeled as the colour of skin
Color.Accordingly, the colour of skin detecting system 1000 of the embodiment of the present invention provides user active correction at this
The mechanism of colour of skin probability statistical table 1070.
In one embodiment, feedback correction module 1040 can be according to checking result corrected parameter correction module
The correction reference of 1030, therefore, can make parameter correction module 1030 raw video is carried out white balance, from
Dynamic exposure, software and hardware parameter etc. are corrected so that the input that colour of skin probability is tabled look-up is the most correct.Citing and
Speech, form access module 1060 according to the rgb value of pixel each in image to colour of skin probability statistical table 1070
Table look-up, to determine that in image, whether each pixel is as the colour of skin, subsequently, produce checking result.This tables look-up
Result can be given feedback correction module 1040 by feedback, and parameter correction module 1030 revises correction reference whereby.
So the pixel representation of the image of the present invention is not limited to RGB pattern, is also optionally used not
Same color space (color space) realizes.
The colour of skin detecting system 1000 being as noted previously, as the embodiment of the present invention contains correction mechanism, logical
Cross automatic correction module 1080, manual correction module 1100, when colour of skin detecting system 1000 finds that the colour of skin is detectd
Sniffing is mistaken, can correct skin tones probability statistical table 1070 immediately, make colour of skin probability statistical table 1070 more add
Whole, and immediately reduce colour of skin detecting False Rate and detecting leak rate, so that the colour of skin of the embodiment of the present invention
The judgement of detecting system 1000 is the most accurate.And can make to input image by feedback correction module 1040
Parameter adjustment is the most correct, in order to make colour of skin detecting system result more accurate.
Embodiment of the present invention colour of skin detecting system 1000 so colour of skin probability statistical table 1070 is carried out school
Just, it is owing to detecting result has detecting erroneous judgement (false positive) and detecting leakage (miss detection),
Cause noise that the colour of skin detects and imperfect.Above-mentioned each correction module can be instant the detecting of the reduction colour of skin by mistake
Sentence rate and detecting leak rate.Furthermore can be according to detecting result manual correction or from dynamic(al) correction with newly-increased or deletion
The colour of skin, and then expand or shrink the scope of the detecting colour of skin.These correct visual practical application request, to continue
Property or intermittent mode are carried out.
In one embodiment, colour of skin detecting system 1000 also comprise colour of skin image sample process module 1110,
Quantization modules 1120 and statistical module 1130.In operation, colour of skin image sample process module 1110
In order to plurality of pictures to be carried out automatic or manual colour of skin labelling;Quantization modules 1120 is in order to automatic or manual
Colour of skin labelling result quantifies;And statistical module 1130 uses statistics or the side of supervised machine learning
Formula calculates the probability that color is the colour of skin of each in aforementioned plurality of pictures, to produce colour of skin probability statistical table
1070。
Specifically, colour of skin image sample process module 1110 carries out the automatic or manual colour of skin to plurality of pictures
Labelling, then made to quantify to the color of color space by quantization modules 1120, selects and a number of represents face
Color.Then, statistical module 1130 uses the mode of statistics or supervised machine learning to calculate each representative
The colour of skin probability of color, if the colour of skin probability come out is more than a threshold value, the most aforementioned color can be judged as
Whether the colour of skin, be then that the result of determination of the colour of skin is fabricated to a form by each color, and aforementioned table is
Colour of skin probability statistical table 1070.
For example, the training method of colour of skin probability statistical table 1070 can be to plurality of pictures automatic or manual
Enclose choosing or labelling skin pixel, then, determine the number of colors number described in statistical table, and usage amount
Change algorithm and determine the representative color of statistical table.Subsequently, the quantified algorithm of each pixel color in picture is calculated
It is which color of statistical table with decision, and record belongs to the colour of skin or the number of times of the non-colour of skin, then come, calculate every
One color belongs to the probability of the colour of skin.Finally, set a threshold value, if the colour of skin probability of arbitrary color is more than this threshold
Value, then be the colour of skin by this color settings, otherwise be then set as the non-colour of skin.Above-mentioned colour of skin probability statistical table 1070
As shown in Table 1.
Table one, colour of skin probability statistical table
As shown in Table 1, the leftmost side is color notation, and C1, C2 and C3 each can represent color space
Three channels, Positives (a) represents that color is registered as the number of times of the colour of skin, Negatives (b) represents color quilt
Being recorded as the number of times of the non-colour of skin, Prob (a/a+b) represents the ratio that color is the colour of skin, and Results (th=0.4) is
Judge that eventually color is whether as the result of the colour of skin.
For example, being numbered the color of 0 in table one and being registered as the number of times of the colour of skin is 50 times, is recorded
Number of times for the non-colour of skin is 100 times, then its be the ratio of the colour of skin be 0.33, here, set a threshold value (th)
Being 0.4, owing to the skin tone of aforementioned color is less than 0.4, the most aforementioned color is judged as the non-colour of skin (No).
The numerical value that right above-mentioned threshold value is used is not limited to the present invention, is familiar with this those skilled in the art and is optionally used
Suitable threshold value realizes the present invention.
In another embodiment, colour of skin detecting system 1000 also comprises image processing module 1090, in order to right
Image carries out image processing to produce image processing result, and provides image processing result to give automatic correction module
To increase newly or to revise from dynamic(al) correction criterion.In operation, image processing module 1090 is in order to carry out image
Image processing can perform mobile pixel and detects producing image processing result, such as this image processing module 1090
Survey, human face detection or other colour of skin detection techniques etc., and provide image processing result to give automatic correction module
The detecting unit 1082 of 1080 calculates, result of calculation be stored in data base 1084 for changing, occuring simultaneously,
Connection collection colour of skin detecting result, and then revise colour of skin probability statistical table 1070 through automatic correction module 1080,
Expand or shrink colour of skin reconnaissance range further.
In an optional embodiment, the automatic correction module 1080 in colour of skin detecting system 1000 is every one
Stabilized image number, such as, every 60 images, will determine that form access module 1060 is given in result transmission,
By form access module 1060 according to judged result to revise colour of skin probability statistical table 1070.So it is not used
To limit the present invention, be familiar with this those skilled in the art obtain optionally use suitable sampling interval to implement the present invention,
Consequently, it is possible to owing to the colour of skin detecting system 1000 of the embodiment of the present invention is able to every stabilized image number
By correction mechanism, colour of skin probability statistical table 1070 is revised, therefore, the skin of the embodiment of the present invention
Color detecting system 1000 can revise colour of skin probability statistical table under the situation of not excessive use system resource
1070, make colour of skin probability statistical table 1070 more complete, so that the colour of skin detecting of the embodiment of the present invention is
The judgement of system 1000 is the most accurate.
In one embodiment, the automatic correction module 1080 of colour of skin detecting system 1000 comprises detecting unit
1082 with data base 1084, wherein data base 1084 comprises from dynamic(al) correction criterion, detecting unit 1082 by
Data base 1084 obtains from dynamic(al) correction criterion, and according to from dynamic(al) correction criterion checking result judged with
Produce judged result.
In another embodiment, colour of skin detecting system 1000 also comprises manual correction module 1100, its in order to
The colour of skin or the non-colour of skin image that are thered is provided according to user increase newly or delete in colour of skin probability statistical table 1070
It is labeled as the color of the colour of skin.Accordingly, at this, offer uses the colour of skin detecting system 1000 of the embodiment of the present invention
The mechanism of person's active correction colour of skin probability statistical table 1070.
Fig. 2 is the schematic flow sheet illustrating a kind of colour of skin method for detecting 200 according to another embodiment of the present invention.
As it can be seen, colour of skin method for detecting comprises the steps of acquirement one image (step 210);According to correction base
It is accurate so that image is carried out parameter correction (step 230);According to the pixel of image, enter to colour of skin probability statistical table
Row tables look-up to produce checking result (step 240);Feedback checking result, and correct according to checking result correction
Benchmark (step 250);Judge to judge knot to produce to checking result from dynamic(al) correction criterion according at least one
Really (step 260);And according to judged result to revise colour of skin probability statistical table (step 270).
Please with reference to Fig. 1 and Fig. 2.In step 210, the step obtaining image can be inputted by image
Module 1010 performs, and aforementioned image input module 1010 refers to any equipment being obtained image, example
Such as camera, network disk, database, the right present invention is not limited thereto.Refer to step 230,
Can by parameter correction module 1030 according to correction reference image to be carried out parameter correction, above-mentioned parameter school
Correction reference is comprised in positive module 1030.For example, if one of parameter set by correction reference is bright
Angle value is 130, and when the brightness value of image is 50, parameter correction module 1030 meeting will according to correction reference
The brightness value of image increases by 80, and makes the brightness value of image reach 130, and the right present invention is not limited thereto,
Parameter correction module 1030 in addition to the brightness value of image, also remaining parameter of recoverable.
In step 240, can be united to colour of skin probability according to the pixel of image by form access module 1060
Meter table 1070 carries out tabling look-up to produce checking result.Additionally, this checking result can be temporarily stored in checking result storage
Storing module 1050, or given parameter correction module 1030 by direct feedback, and by parameter correction module 1030
According to checking result correction correction reference.
For example, can by form access module 1060 according to the rgb value of pixel each in image come to
Colour of skin probability statistical table 1070 is tabled look-up, to determine that in image, whether each pixel is as the colour of skin, subsequently,
Produce checking result.So the pixel representation of the image of the present invention is not limited to RGB pattern, also may be used
Different color space (color space) is optionally used to realize.
Refer to step 260, can by automatic correction module 1080 according at least one from dynamic(al) correction criterion pair
Checking result carries out judging to produce judged result, and can pass through form access module 1060 in step 270
According to judged result to revise colour of skin probability statistical table 1070.For example, aforementioned can from dynamic(al) correction criterion
For isolated point (isolated point) determining method, and by automatic correction module 1080 according to isolated point determining method
The mode judging checking result is recorded in the related description of Fig. 1 the most in detail, does not goes to live in the household of one's in-laws on getting married at this
State.
The colour of skin method for detecting 200 being as noted previously, as the embodiment of the present invention is achieved correction mechanism, logical
Cross this correction mechanism, when colour of skin method for detecting 200 running finds mistake, can immediately revise aforementioned mistake,
And synchronize to revise colour of skin probability statistical table 1070, make colour of skin probability statistical table 1070 more complete, so that
The judgement of the colour of skin method for detecting 200 of the embodiment of the present invention is the most accurate.
Why colour of skin method for detecting 200 disclosed by the embodiment of the present invention will be to colour of skin probability statistical table
1070 are corrected, and are owing to detecting result has detecting erroneous judgement (false positive) and detecting leakage
(miss detection), causes noise that the colour of skin detects and imperfect.Above-mentioned each correction module can be instant
The reduction colour of skin detecting False Rate and detecting leak rate.Furthermore can be according to detecting result feedback correction input picture
Face and manual correction or from dynamic(al) correction with newly-increased or delete the colour of skin, and then expand or shrink the model of the detecting colour of skin
Enclose.These corrections can all be carried out by every picture, or intermittent implementation.
Above-mentioned automatic correcting mode can be realized by a series of detecting unit 182, and it is in order to automatically to calculate
False skin pixel (false positive) and false non-skin pixel (miss detection), access mould through form
Block 1060 corrects (such as: newly-increased or deletion) colour of skin probability statistical table.The sense mode of detecting unit 182
It is recorded in the most in detail in the explanation of Fig. 1, does not repeats at this.
In one embodiment, colour of skin method for detecting 200 also comprises the steps of and image is carried out image processing
To produce image processing result, and according to image processing result to revise colour of skin probability statistical table 1070 (step
280)。
In step 280, image is carried out image processing to produce image processing result, and provide at image
Reason result gives automatic correction module to increase newly or to revise from dynamic(al) correction criterion.This step can pass through image processing mould
Block 1090 performs, and for example, image processing module 1090 in order to carry out image processing to produce to image
Raw image processing result, such as this image processing module 1090 can perform mobile pixel detecting, human face detection
Or other colour of skin detection techniques etc., and provide image processing result to give the detecting list of automatic correction module 1080
Unit 1082 calculates, and result of calculation is stored in data base 1084 for changing, occur simultaneously, join collection colour of skin detecting knot
Really, and then revise colour of skin probability statistical table 1070 through automatic correction module 1080, expand further or contracting
Little colour of skin reconnaissance range.
In the step 270 of colour of skin method for detecting 200, according to judged result to revise colour of skin probability statistical table
Step, can by automatic correction module 1080 every a stabilized image number, such as every 60 images,
Will determine that form access module 1060 is given in result transmission, so that form access module 1060 is correspondingly every one
Stabilized image number, according to judged result to revise colour of skin probability statistical table 1070.So itself and be not used to limit
Determine the present invention, be familiar with this those skilled in the art and obtain and optionally use suitable sampling interval to implement the present invention.
Consequently, it is possible to owing to the colour of skin method for detecting 200 of the embodiment of the present invention is able to every stabilized image number
Just by correction mechanism, colour of skin probability statistical table 1070 is revised, therefore, the embodiment of the present invention
Colour of skin method for detecting 200 can revise colour of skin probability statistical table under the situation of not excessive use system resource
1070, make colour of skin probability statistical table 1070 more complete, so that the colour of skin detecting side of the embodiment of the present invention
The judgement of method 200 is the most accurate.
Fig. 3 is to illustrate colour of skin probability statistical table in a kind of colour of skin method for detecting according to further embodiment of this invention
Production method 300 flow chart.The production method 300 of aforementioned colour of skin probability statistical table comprises the steps of
Plurality of pictures is carried out automatic or manual colour of skin labelling (step 310);Make to quantify to the color of color space,
Select and a number of represent color.(step 320);The mode using statistics or supervised machine learning is come
Calculate the probability (step 330) that color is the colour of skin of each in aforementioned plurality of pictures;And produce colour of skin machine
Rate statistical table (step 340).
Refer to step 310, by colour of skin image sample process module 1110, plurality of pictures can be carried out certainly
Dynamic or manual colour of skin labelling, then, in step 320, the automatic or manual colour of skin to aforementioned plurality of pictures
The step that labelling result carries out quantifying can be performed by quantization modules 1120, subsequently, in step 330,
The probability that color is the colour of skin of each in aforementioned plurality of pictures can be added up by statistical module 1130, then
Person, in step 340, can produce colour of skin probability statistical table 1070 by statistical module 1130.
Specifically, the colour of skin image sample process module 1110 skin-color training sample to collecting can be passed through
(aforementioned plurality of pictures) carries out automatic or manual colour of skin labelling, then by quantization modules 1120 to skin-color training
The automatic or manual colour of skin labelling result of sample quantifies, and then, uses system by statistical module 1130
The mode of meter or supervised machine learning calculates the colour of skin probability of each color, if the colour of skin come out
Whether probability is more than a threshold value, and the most aforementioned color can be judged as the colour of skin, be then the colour of skin by each color
Result of determination is fabricated to a form, and colour of skin probability statistical table 1070 is i.e. produced by above step.
In another embodiment, colour of skin method for detecting 200 also comprise the steps of according to input the colour of skin with
And non-colour of skin sample is with manual correction colour of skin probability statistical table.In this step, can manually correction module
1100 colour of skin or the non-colour of skin images provided according to user increase or delete colour of skin color newly so that skin
The skin pixel that color probability statistical table is recorded can dynamically adjust.Accordingly, the colour of skin of the embodiment of the present invention is detectd
Survey method 200 provides the mechanism of user active correction colour of skin probability statistical table 1070 at this.
Colour of skin method for detecting 200 as above all can be performed by software, hardware and/or a piece of wood serving as a brake to halt a carriage body.Citing
For, if to perform speed and accuracy for primarily considering, the most substantially can be selected for hardware and/or a piece of wood serving as a brake to halt a carriage body be
Main;If with design flexibility for primarily considering, it is main for the most substantially can be selected for software;Or, can use simultaneously
Software, hardware and a piece of wood serving as a brake to halt a carriage body work compound.It will be understood that these examples provided above do not have what is called, who is excellent
Who bad point, also and be not used to limit the present invention, those who are familiar with this art needed elastic design at that time when regarding.
Furthermore, art has usually intellectual when it can be appreciated that in colour of skin method for detecting 200
Each step is named, merely to it is understandable to allow the technology of this case become apparent from, not according to its function performed
In order to limit such step.Each step is integrated into same step or is split into multiple step, or will
Either step changes to perform in another step, all still falls within embodiments of the present invention.
From the invention described above embodiment, the application present invention has following advantages.The embodiment of the present invention carries
For a kind of colour of skin detecting system 1000 and colour of skin method for detecting 200, by automatic correction module 1080 and
Manual correction module 1100, when colour of skin detecting system 1000 and method 200 find colour of skin detecting mistake,
Can correct skin tones probability statistical table 1070 immediately, make colour of skin probability statistical table 1070 more complete, and immediately
Ground reduces colour of skin detecting False Rate and detecting leak rate, so that the colour of skin detecting system of the embodiment of the present invention
1000 and the judgement of method 200 the most accurate, furthermore it is possible to according to detecting result feedback correction input
Picture makes look-up table the most correct.
Additionally, due to the colour of skin detecting system 1000 of the embodiment of the present invention and method 200 are able to every solid
Colour of skin probability statistical table 1070 is just revised by fixing number by correction mechanism, therefore, and this
The colour of skin detecting system 1000 of bright embodiment and method 200 can be in the situations of not excessive use system resource
Under, revise colour of skin probability statistical table 1070.
Although the present invention is disclosed above with embodiment, so it is not limited to the present invention, any familiar
This those skilled in the art, without departing from the spirit and scope of the present invention, when being used for a variety of modifications and variations, therefore
Protection scope of the present invention is when being defined in the range of standard depending on appending claims.
Claims (12)
1. a colour of skin method for detecting, it is characterised in that comprise:
Obtain an image;
According to the pixel of this image, carry out tabling look-up to produce a checking result to a colour of skin probability statistical table;
Judge to produce a judged result to this checking result from dynamic(al) correction criterion according at least one;And
According to this judged result to revise this colour of skin probability statistical table,
Wherein should comprise from the automatic correcting mode of dynamic(al) correction criterion that automatically to calculate the pixel of this image be false skin
Color pixel or false non-skin pixel.
Colour of skin method for detecting the most according to claim 1, it is characterised in that also comprise:
According to a correction reference so that this image is carried out parameter correction;And
This checking result of feedback, and according to this this correction reference of checking result correction.
Colour of skin method for detecting the most according to claim 1, it is characterised in that also comprise:
Plurality of pictures is carried out automatic or manual colour of skin labelling;
This automatic or manual colour of skin labelling result is quantified;And
The mode using statistics or supervised machine learning calculates the color of each in described plurality of pictures
For the probability of the colour of skin, to produce this colour of skin probability statistical table.
Colour of skin method for detecting the most according to claim 1, it is characterised in that also comprise:
This image is carried out image processing to produce an image processing result, and according to this image processing result with
Newly-increased or correction should be from dynamic(al) correction criterion.
Colour of skin method for detecting the most according to claim 1, it is characterised in that this is tied according to this judgement
Fruit, to revise the step of this colour of skin probability statistical table, is to perform once every a stabilized image number.
Colour of skin method for detecting the most according to claim 1, it is characterised in that also comprise:
The colour of skin and non-colour of skin sample according to inputting are with this colour of skin probability statistical table of manual correction.
7. a colour of skin detecting system, it is characterised in that comprise:
One image input module, in order to obtain an image;
One form access module, in order to the pixel according to this image, tables look-up to a colour of skin probability statistical table
To produce a checking result;And
One automatic correction module, in order to basis at least from dynamic(al) correction criterion, judges this checking result
To produce a judged result, and this judged result transmission is given this form access module, wherein should be from dynamic(al) correction
The automatic correcting mode of criterion comprises that automatically to calculate the pixel of this image be false skin pixel or false non-colour of skin picture
Element, and this form access module is according to this judged result, to revise this colour of skin probability statistical table.
Colour of skin detecting system the most according to claim 7, it is characterised in that also comprise:
One parameter correction module, comprises a correction reference, and wherein this parameter correction module is according to this correction base
It is accurate so that this image is carried out parameter correction;And
One feedback correction module, in order to be obtained this checking result, and this knot of tabling look-up of feedback by form access module
Fruit gives this parameter correction module, and wherein this parameter correction module can be according to this correction base of this checking result correction
Accurate.
Colour of skin detecting system the most according to claim 7, it is characterised in that also comprise:
One colour of skin image sample process module, in order to carry out automatic or manual colour of skin labelling to plurality of pictures;
One quantization modules, in order to quantify this automatic or manual colour of skin labelling result;And
One statistical module, uses the mode of statistics or supervised machine learning to calculate in described plurality of pictures every
The color of one is the probability of the colour of skin, to produce this colour of skin probability statistical table.
Colour of skin detecting system the most according to claim 7, it is characterised in that also comprise:
One image processing module, in order to this image is carried out image processing to produce an image processing result, and
It is somebody's turn to do from dynamic(al) correction criterion with newly-increased or correction according to this image processing result.
11. colour of skin detecting systems according to claim 7, it is characterised in that this automatic correction module
Will determine that this form access module is given in result transmission every a stabilized image number, by this form access module root
According to this judged result to revise this colour of skin probability statistical table.
12. colour of skin detecting systems according to claim 7, it is characterised in that also comprise:
One manual correction module, in order to come new according to the colour of skin or non-colour of skin image that input this colour of skin detecting system
Increase or delete the color being labeled as the colour of skin in this colour of skin probability statistical table.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW100149977A TWI537877B (en) | 2011-12-30 | 2011-12-30 | Method for detecting skin tone and skin tone detection system |
TW100149977 | 2011-12-30 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103186785A CN103186785A (en) | 2013-07-03 |
CN103186785B true CN103186785B (en) | 2016-10-26 |
Family
ID=48677945
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210331409.2A Active CN103186785B (en) | 2011-12-30 | 2012-09-07 | Skin color detection method and system |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN103186785B (en) |
TW (1) | TWI537877B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109975299B (en) * | 2017-12-27 | 2022-03-25 | 群光电子股份有限公司 | Light source detection system and method |
WO2020164820A1 (en) | 2019-02-14 | 2020-08-20 | International Business Machines Corporation | Directed interrupt virtualization |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6072526A (en) * | 1990-10-15 | 2000-06-06 | Minolta Co., Ltd. | Image sensing device that can correct colors corresponding to skin in a video signal |
CN101251898A (en) * | 2008-03-25 | 2008-08-27 | 腾讯科技(深圳)有限公司 | Skin color detection method and apparatus |
TW201100058A (en) * | 2009-06-17 | 2011-01-01 | Univ Southern Taiwan | System and method of using digital image multi-region analysis to establish and analyze skin parameters |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101549824B1 (en) * | 2008-12-11 | 2015-09-03 | 삼성전자주식회사 | Method and apparatus for correcting skin color and digital photographing apparatus using thereof |
-
2011
- 2011-12-30 TW TW100149977A patent/TWI537877B/en active
-
2012
- 2012-09-07 CN CN201210331409.2A patent/CN103186785B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6072526A (en) * | 1990-10-15 | 2000-06-06 | Minolta Co., Ltd. | Image sensing device that can correct colors corresponding to skin in a video signal |
CN101251898A (en) * | 2008-03-25 | 2008-08-27 | 腾讯科技(深圳)有限公司 | Skin color detection method and apparatus |
TW201100058A (en) * | 2009-06-17 | 2011-01-01 | Univ Southern Taiwan | System and method of using digital image multi-region analysis to establish and analyze skin parameters |
Also Published As
Publication number | Publication date |
---|---|
TW201327475A (en) | 2013-07-01 |
TWI537877B (en) | 2016-06-11 |
CN103186785A (en) | 2013-07-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20150304525A1 (en) | Color correction based on multiple images | |
US20180322367A1 (en) | Image processing method, non-transitory computer readable storage medium and image processing system | |
WO2021189847A1 (en) | Training method, apparatus and device based on image classification model, and storage medium | |
US9712796B2 (en) | Metadata for use in color grading | |
CN109064525A (en) | Picture format conversion method, device, equipment and storage medium | |
CN106530361A (en) | Color correction method for color face image | |
CN110766711A (en) | Video shot segmentation method, system, device and storage medium | |
CN110910401A (en) | Semi-automatic image segmentation data annotation method, electronic device and storage medium | |
CN110990617B (en) | Picture marking method, device, equipment and storage medium | |
CN113837191A (en) | Cross-satellite remote sensing image semantic segmentation method based on bidirectional unsupervised domain adaptive fusion | |
US20200282811A1 (en) | Adaptive glare removal and/or color correction | |
CN113538223A (en) | Noise image generation method, noise image generation device, electronic device, and storage medium | |
CN103186785B (en) | Skin color detection method and system | |
CN113920378B (en) | Bupleurum seed identification method based on attention mechanism | |
CN105893938A (en) | Oil temperature gauge reading method and system | |
CN114758054A (en) | Light spot adding method, device, equipment and storage medium | |
CN110751668A (en) | Image processing method, device, terminal, electronic equipment and readable storage medium | |
CN106454509B (en) | A kind of detection method and device of advertising image | |
CN113628259A (en) | Image registration processing method and device | |
CN106375849A (en) | Template generation method, template generation device, video updating method and video updating device | |
CN107230195A (en) | A kind of image treatment method and device | |
CN105893943A (en) | Oil level detection method and system | |
CN104050455B (en) | A kind of skin color detection method and system | |
KR102215607B1 (en) | Electronic device capable of correction to improve the brightness of dark images and operating method thereof | |
CN111898525A (en) | Smoke recognition model construction method, smoke detection method and smoke detection device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |