CN109461118A

CN109461118A - A kind of image processing method and device

Info

Publication number: CN109461118A
Application number: CN201811339208.0A
Authority: CN
Inventors: 高超; 王通; 郑柏亭
Original assignee: Taipu Intelligent Co ltd
Current assignee: Beijing Ihandy Mobile Internet Technology Co ltd
Priority date: 2018-11-12
Filing date: 2018-11-12
Publication date: 2019-03-12

Abstract

The present invention provides a kind of image processing method and device, obtain personage's picture to be processed, the first picture of each target site is extracted from personage's picture to be processed, to either objective position: using the first picture of the target site as the input of the corresponding default identification model of the target site, obtain the mark of the target site of the default identification model output, mark based on the target site, the second picture of the target site shown in a special way is obtained from default picture library, and determine the skin color of personage's picture to be processed, hair color and pupil color, second picture based on each target site, skin color, hair color and pupil color, obtain the personage's picture shown with special image, personage's picture to be processed is automatically converted to the personage's picture shown with special image to realize, and which can So that user's need not have fine arts grounding in basic skills can also obtain the special image for meeting user's esthetic requirement while the complexity that can reduce conversion.

Description

A kind of image processing method and device

Technical field

The invention belongs to image processing technologies, more specifically more particularly to a kind of image processing method and device.

Background technique

It, can be by having the APP (Application, using journey for repairing figure function in terminal at present for equal pictures of taking pictures certainly Sequence) it mixed colours picture, increase special efficacy and the specially treateds such as thin face are carried out to the personage in picture, or for a cartoon Personage modifies five official ranks of cartoon figure by having the APP for repairing figure function, but has the APP for repairing figure function cannot at present Convert the face with special image for the face in personage's picture, the face in personage's picture cannot be such as automatically synthesized for Face with cartoon character.

Summary of the invention

In view of this, the purpose of the present invention is to provide a kind of image processing method and device, for realizing will be to be processed Personage's picture is converted to the personage's picture shown with special image.Technical solution is as follows:

The present invention provides a kind of image processing method, comprising:

Obtain personage's picture to be processed；

The first picture of each target site, each target site packet are extracted from personage's picture to be processed It includes: at least one position in nose, mouth, eyes, ear, eyebrow and hair；

To the either objective position in each target site: the first picture of the target site is determined as the target The input of the corresponding default identification model in position obtains the target site of the corresponding default identification model output of the target site Mark, wherein the corresponding default identification model of the target site is by the of the target site extracted from historical personage's picture One picture is trained to obtain as input using the mark marked for the target site as output；

To the either objective position in each target site: the mark based on the target site is obtained from default picture library The second picture of the target site is obtained, wherein the second picture of the target site shows the target site with special image；

Determine the skin color of personage's picture to be processed, the hair color of personage's picture to be processed and it is described to Handle the pupil color of personage's picture；

Second picture, the skin color of personage's picture to be processed, the people to be processed based on each target site The pupil color of the hair color of object picture and personage's picture to be processed obtains the personage's picture shown with special image.

Preferably, the skin color of determination personage's picture to be processed includes:

Determine that the skin area in personage's picture to be processed, the skin area are personage's pictures to be processed Meet the region of pre-set color condition in face and/or face's following region；

Interference of the filtering environmental factor to the skin color in the skin area obtains personage's picture to be processed Skin color.

Preferably, the default identification model exports number of the target site in the default picture library, by institute The mark that number of the target site in the default picture library is determined as the target site is stated, and the same number corresponds to It include an at least width second picture in the default picture library, the phase between each width picture in an at least width second picture Like degree in default similarity dimensions；

The either objective position in each target site: the mark based on the target site, from default picture library The middle second picture for obtaining the target site includes: to determine the target site in the default picture in the default picture library The corresponding each width second picture of number in library, chooses a width picture from identified each width second picture.

Preferably, the second picture based on each target site, the skin color of personage's picture to be processed, institute The hair color of personage's picture to be processed and the pupil color of personage's picture to be processed are stated, is obtained with special image displaying Personage's picture:

Determine the second picture position and corresponding direction of each target site；

Second picture position and corresponding direction based on each target site, to the second figure of each target site Piece is spliced, and the personage's picture shown with special image is obtained；

Set described wait locate for skin color, hair color and the pupil color of the personage's picture shown with special image Manage skin color, hair color and the pupil color of personage's picture.

Preferably, the method also includes:

User is obtained to the feedback information of personage's picture with feature image display, the feedback information is at least used for Show the user to the satisfaction of the face picture shown with special image；

Based on the feedback information, the default identification model is corrected again, to modify the default identification mould Corresponding relationship between the mark of the target site of the input of type and the default identification model output.

The present invention also provides a kind of picture processing units, comprising:

Module is obtained, for obtaining personage's picture to be processed；

Extraction module, it is described for extracting the first picture of each target site from personage's picture to be processed Each target site includes: at least one position in nose, mouth, eyes, ear, eyebrow and hair；

Identification module, for the either objective position in each target site: by the first figure of the target site Piece is determined as the input of the corresponding default identification model of the target site, and it is defeated to obtain the corresponding default identification model of the target site The mark of the target site out, wherein the target site corresponding default identification model will be extracted from historical personage's picture First picture of the target site is trained to obtain as input using the mark marked for the target site as output；

Second picture determining module, for the either objective position in each target site: based on the target site Mark, obtains the second picture of the target site, wherein the second picture of the target site is with special form from default picture library As showing the target site；

Color determination module, for determining skin color, the personage's picture to be processed of personage's picture to be processed Hair color and personage's picture to be processed pupil color

Synthesis module, for second picture, personage's picture to be processed based on each target site skin color, The pupil color of the hair color of personage's picture to be processed and personage's picture to be processed, obtains showing with special image Personage's picture.

Preferably, the color determination module, specifically for the skin area in determination personage's picture to be processed, and Interference of the filtering environmental factor to the skin color in the skin area obtains the skin face of personage's picture to be processed Color, the skin area be personage's picture to be processed face and/or face's following region in meet pre-set color condition Region.

Preferably, the identification module, for exporting number of the target site in the default picture library, by institute The mark that number of the target site in the default picture library is determined as the target site is stated, and the same number corresponds to At least width second picture for including in the default picture library, between each width picture in an at least width second picture Similarity is in default similarity dimensions；

The second picture determining module, for determining the target site in the default figure in the default picture library The corresponding each width second picture of the number of valut, chooses a width picture from identified each width second picture.

Preferably, the synthesis module comprises determining that unit, concatenation unit and synthesis unit,

The determination unit, for determining the second picture position and corresponding direction of each target site；

The concatenation unit, for second picture position and corresponding direction based on each target site, to each The second picture of a target site is spliced, and the personage's picture shown with special image is obtained；

The synthesis unit, skin color, hair color and the pupil of personage's picture for will be shown with special image Color is set as skin color, hair color and the pupil color of personage's picture to be processed.

Preferably, described device further includes correction module, and the correction module includes: obtaining unit and amending unit,

The acquiring unit, for obtaining user to the feedback information of personage's picture with feature image display, institute Stating feedback information is at least used to show the user to the satisfaction of the face picture shown with special image；

The amending unit is corrected the default identification model, again for being based on the feedback information to repair Change the corresponding relationship between the mark of the target site of input and the default identification model output of the default identification model.

From above-mentioned technical proposal it is found that being mentioned from personage's picture to be processed in the case where obtaining personage's picture to be processed The first picture for taking out each target site, to the either objective position in each target site: by the first of the target site Picture is determined as the input of the corresponding default identification model of the target site, obtains the corresponding default identification model of the target site The mark of the target site of output, the mark based on the target site obtain the of the target site from default picture library Two pictures, wherein the second picture of the target site shows the target site with special image；And determining figure map to be processed In the case where the pupil color of the skin color of piece, the hair color of personage's picture to be processed and personage's picture to be processed, it is based on Skin color, hair color and the pupil color of the second picture of each target site and personage's picture to be processed, obtain with spy Personage's picture to be processed is automatically converted to the figure map shown with special image to realize by personage's picture of different image display Piece, and above-mentioned transform mode can choose second picture from default picture library automatically, so that user's need not have fine arts function Bottom can also obtain the special image for meeting user's esthetic requirement, and above-mentioned conversion is selected from default picture library manually without user Interested second picture is taken, so as to reduce the complexity of conversion.

Detailed description of the invention

In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is the present invention Some embodiments for those of ordinary skill in the art without creative efforts, can also basis These attached drawings obtain other attached drawings.

Fig. 1 is a kind of flow chart of image processing method provided in an embodiment of the present invention；

Fig. 2 is the schematic diagram of default identification model training provided in an embodiment of the present invention；

Fig. 3 is the schematic diagram of image processing method provided in an embodiment of the present invention；

Fig. 4 is the flow chart of another image processing method provided in an embodiment of the present invention；

Fig. 5 is a kind of structural schematic diagram of picture processing unit provided in an embodiment of the present invention；

Fig. 6 is the structural schematic diagram of synthesis module in picture processing unit provided in an embodiment of the present invention；

Fig. 7 is the structural schematic diagram of another picture processing unit provided in an embodiment of the present invention；

Fig. 8 is the structural schematic diagram of correction module in picture processing unit provided in an embodiment of the present invention.

Specific embodiment

In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.

Referring to Fig. 1, it illustrates a kind of flow chart of image processing method provided in an embodiment of the present invention, for will be to Processing personage's picture is automatically converted to the personage's picture shown with special image, specifically, image processing method shown in Fig. 1 can be with The following steps are included:

S101 obtains personage's picture to be processed, it can be understood as: the picture with personage can be considered as personage to be processed Picture, such as user take pictures certainly or the picture including personage and landscape, and the mode for obtaining personage's picture to be processed can Be user upload picture, from the picture that other equipment receive and from least one in the picture downloaded in other equipment Kind.

S102, extracts the first picture of each target site from personage's picture to be processed, and each target site includes: At least one position in nose, mouth, eyes, ear, eyebrow and hair, that is to say, that mentioned from personage's picture to be processed It takes one or more target sites, such as personage's picture to be processed to can be the picture including target complete position, is also possible to The mesh that picture including partial target position, then the first picture of each target site extracted and personage's picture to be processed include It is related to mark position.

It, will be from personage to be processed such as if personage's picture to be processed is the picture for including target complete position The first picture that each target site is extracted in picture is respectively: the first picture, the eyes of the first picture of nose, mouth First picture, the first picture of ear, the first picture of eyebrow and hair the first picture, if personage's picture to be processed be packet The picture at partial target position is included, specific personage's picture to be processed can only include eyes and eyebrow target site, then extract The first picture of each target site only has out: the first picture of eyebrow and the first picture of eyes.

In the present embodiment, one kind that the first picture of each target site is extracted from personage's picture to be processed is feasible Mode is: carrying out recognition of face to figure map's piece to be processed, identification is obtained nose, mouth, eyes, ear, eyebrow and hair In at least one position, by nose, mouth, eyes, ear, eyebrow and hair at least one position region carry out It cuts, target site is separated from personage's picture to be processed, to obtain nose, mouth, eyes, ear, eyebrow With first picture at the either objective position of hair, such as multiple keys of above-mentioned each target site are obtained by recognition of face Point, to either objective position in each target site: determining target site place by multiple key points of the target site Then determined region is extracted from personage's picture to be processed, obtains the first picture of the target site by region.

Another feasible pattern for extracting the first picture of each target site is: training obtains picture and extracts mould in advance Personage's picture to be processed is input to the picture and extracted in model by type, obtains each target portion that the picture extracts model output First picture of position, can also train in advance in practical applications certainly and obtain multiple pictures extraction models, and each picture extracts Model corresponds to a target site, i.e. a picture extracts the first picture that model is used to extract a target site, wherein in advance The process that first training obtains picture extraction model may is that be determined from default personage's picture set by cross validation mode Training set and test set, to any personage's picture in training set: marking out each target portion from personage's picture Position, each target site based on mark carry out model training, obtain picture and extract model, are obtained with extracting model by picture The key point of each target site, and model is extracted by picture, each target portion is exported based on the key point of each target site Then first picture of position extracts model to obtained picture by each personage's picture in test set again and tests, It determines that picture extracts model and can extract the first picture of each target site if testing, terminates to train, otherwise again to instruction Personage's picture in white silk set is labeled is trained again.

Additionally need to illustrate a bit: picture, which extracts model, can be directed at least one people, that is to say, that a picture mentions Modulus type can be only applicable to a people, obtain picture by personage's picture training of a people and extract model or a figure Piece, which extracts model, can be adapted for multiple people, obtains picture by personage's picture training of multiple people and extracts model.

S103, to the either objective position in each target site: using the first picture of the target site as the target The input of the corresponding default identification model in position obtains the target site of the corresponding default identification model output of the target site Mark, wherein the corresponding default identification model of the target site is by the of the target site extracted from historical personage's picture One picture is trained to obtain as input using the mark marked for the target site as output.That is, passing through face Historical personage's picture training obtain each target site respectively corresponding default identification model, and then subsequent can be applied to face Personage's picture to be processed mark identification.

In the present embodiment, the acquisition process of identification model is preset are as follows: the set for obtaining historical personage's picture of face, from Training set and test set are marked off in the set of historical personage's picture, and model is extracted by recognition of face or above-mentioned picture The first picture of a target site in any historical personage's picture in training set is obtained, and marks out a target site Mark carries out model training with the first picture of the target site and the mark marked out, obtains corresponding default identification model, Wherein the obtained default identification model of training is corresponding with target site respectively, that is to say, that each target site has pair The default identification model answered, if nose target site has corresponding default identification model, the nose target site got First picture will be input to the corresponding default identification model of nose target site；Then pass through each going through in test set History personage picture tests default identification model, to determine whether default identification model meets the requirements, undesirable In the case where can realize optimization to default identification model by repetition training and test, can be according to reality for desired setting Depending on the application of border, this present embodiment is no longer illustrated.

It in practical annotation process, needs to classify to either objective position, and is directed to for either objective position, it is right The special image of face and the shape of user in the thinner personage's picture that can to show with special image of the classification of the target site As being more bonded, such as by taking the hair of women as an example, the hair of women can be divided into but be not limited to long hair, middle hair, bob and super Bob can be divided into big wave curly hair, bubble roll hair and straight hair in long hair again, then need to distinguish for these types in assorting process A mark is marked out, it is consistent or close with the feature having by the corresponding type of the same mark, but different identification The distinctiveness for the feature that corresponding type has becomes apparent from, and can make same text in datagram valut by the classification in this way (the first picture in such as same file has the similarity between the first picture in part folder in default similarity dimensions Feature it is consistent or close), the higher the better for the degree of isolation of the first picture of different files, in this way based on default identification mould Type can make the mark of target site under different files different when identifying.Still by taking the hair of women as an example, No. 00 file In the hair styles of hair of all first pictures be all consistent as far as possible, and the hair style in the first picture in No. 01 file It will be with the difference of the hair style in the first picture in No. 00 file as far as possible, then identifying the two by default identification model The mark of the hair obtained when hair under file is different, so that second picture also can be different, realizes to different target Position shows the second picture of different special images.

Wherein No. 00 and No. 01 can then be considered as mark of this target site of hair in data picture library, have same The default similarity of all first pictures of one mark in default similarity dimensions, all first pictures of different identification it is pre- If similarity should be less than default similarity dimensions, meanwhile, the quantity of the first picture with like-identified is greater than or equal to 1, Being exactly includes at least one secondary first picture in certain mark, includes at least one first in file such as by taking file as an example Picture, and the similarity between each the first picture in same file folder is being preset in similarity dimensions, different files The default similarity between each the first picture in folder is not in default similarity dimensions.

In the present embodiment, can be using such as to the classification of target site: AlexNet, VGGNET16, GoogleNet, The picture classifications mode such as SimpleNet, the classification accuracy obtained by these picture classification modes are as shown in table 1 below:

The classification accuracy of 1 picture classification mode of table

Mode	Memory size (M)	T1/T5 accuracy rate
			AlexNet	60	57.2/80.3
VGGNet16	138	70.5
			GoogleNet	8	68.7
WideResNet	11.7	69.6/89.07
			SimpleNet	5.4	67.17/87.44

Needing explanation herein is a bit: at the terminal due to image processing method provided in this embodiment operation, and eventually The memory at end is smaller, it is therefore desirable to be classified using a kind of lesser picture classification mode of memory requirements, and 1 institute of consolidated statement The classification accuracy shown is it is found that it is preferred that SimpleNet.Furthermore the operation of SimpleNet is simple, and off-line operation may be implemented, and makes The personage's picture shown with special image can be obtained by image processing method offline by obtaining, and so as to improve recognition speed and be improved User experience.

It combines Fig. 2 to be illustrated default identification model by taking eyes as an example below, is concentrated from eyes labeled data and choose one A picture exports the picture of eyes in selected picture by picture pretreatment (model is extracted in such as recognition of face or above-mentioned picture) (i.e. the first pictures of eyes) input the mark (marks of such as eyes) inside the picture of the eyes and eyes labeled data collection Into model training code, eye model is exported.It in the present embodiment, can be with to the either objective position of above-mentioned target site Training obtains a default identification model, and carries out fine cutting to figure map's piece to be processed, obtains either objective position First picture, to be identified by respective default identification model, so that the influence of other target sites is reduced, so that identification It is more accurate.

S104, to the either objective position in each target site: the mark based on the target site, from default picture library The middle second picture for obtaining the target site, wherein the second picture of the target site shows the target site with special image.

In this embodiment, the second picture preset in picture library can be by the personnel with fine arts grounding in basic skills according to target site Reality image be designed so that the target site that is shown of second picture has special image, such as cartoon character, ancient costume At least one of image, soldier's image etc. image.

Target site is identified as presetting the number in picture library by target site, is obtained from default picture library A kind of feasible pattern of the second picture of the target site is: determining the target site in default picture library in default picture library In the corresponding each width second picture of number, choose a width picture from identified each width second picture, specific default figure The same number corresponds to a file in valut, can have one or more second picture under a file, according to each Number of the target site in default picture library obtains a file, and then determines a width second picture from this document folder As the second picture of the target site, it can such as randomly select a width or be chosen according to the priority of picture, wherein picture Priority can be depending on the number that picture is selected, specific the present embodiment no longer illustrates.For nose, existed based on nose Number in default picture library is obtained with special image from default picture library, as cartoon character shows the second picture of nose.

The similarity between each width second picture under a file is being preset in similarity dimensions in the present embodiment, With guarantee same file folder under each width second picture between it is similar to each other and from each width second picture under different files Difference, so that obtained second picture different from can be numbered with based on other by numbering obtained second picture based on one.

S105 determines the skin color of personage's picture to be processed, the hair color of personage's picture to be processed and people to be processed The pupil color of object picture.In the present embodiment, determining a kind of feasible pattern of the skin color of personage's picture to be processed is: really Skin area in fixed personage's picture to be processed, wherein skin area is the face and/or face of personage's picture to be processed or less Meet the region of pre-set color condition in region；Interference of the filtering environmental factor to the skin color in skin area, obtain to Handle the skin color of personage's picture.

Such as it determines that the implementation of the skin area in personage's picture to be processed may is that and determines figure map to be processed Face area and face's following region in piece, to RGB (Red-Green-Blue, the RGB) color in this two parts region Spatial dimension is reduced, and obtains meeting the region of pre-set color condition as skin area, and wherein pre-set color condition can be with It is following but is not limited to following formula:

R > 95AndG > 40AndB > 20AndR > BAndMax (R, G, B)-Min (R, G, B) > 15AndAbs (R-G) > 15

Wherein, R indicates red value, and G indicates the value of green, and B indicates the value of blue, and the value of R, G, B Range is 0-255, And indicate and logical relation, Max (R, G, B) expression takes R, the maximum value in G, B, Min (R, G, B) Expression takes R, the minimum value in G, B, and Abs (R-G) expression takes R-G to obtain absolute value.

And interference of the filtering environmental factor to the skin color in skin area, obtain the skin face of personage's picture to be processed Color implementation may is that the color by all pixels point in skin area is weighted summation, obtains mean skin color, And carry out eliminating the processing of environmental disturbances factor, obtain skin area color.

Such as it carries out elimination environmental disturbances factor processing implementation and may is that the mean skin color that will be indicated with RGB It is converted into HSB (Hues-Saturation-Brightness, hue saturation brightness) expression, and obtained HSB is adjusted It is whole, adjust S component (S in HSB, saturation degree) and B component (B in HSB, brightness), specific adjustment mode is as follows: judgement Whether B component belongs to the first preset range, and B component and S component increase by the first preset increments value if belonging to, if be not belonging to Then continuing to judge whether B component belongs to the second preset range, B component and S component increase by the second preset increments value if belonging to, B component and S component increase third preset increments value if being not belonging to, it should be noted that B component and S component are by increasing After preset increments value, the final B component and S component that should ensure that are no more than 1.Wherein, the first preset range, second pre- If range, the first preset increments and the second preset increments can depend on the circumstances, B component such as is judged whether less than 0.5, if met B component is less than 0.5, then B component increases by 0.5 and S component and increases by 0.3, if not meeting B component less than 0.5, continues to judge B Whether component is less than 0.7, if meeting B component less than 0.7, B component increases by 0.15 and S component and increases by 0.15 if category, if not Meet B component less than 0.7, then B component increases by 0.25 and S component and increases by 0.25, it should be noted that B component and S component pass through After increasing preset increments value, the final B component and S component that should ensure that are no more than 1.About preset range and default increasing Other numerical value setting of amount, this implementation are not illustrating.

Determining a kind of feasible mode of the hair color of personage's picture to be processed is: extracting to from personage's picture to be processed Hair the first picture, pixel in the first picture of hair is handled, hair color, such as first to hair are obtained The color of all pixels point in picture is weighted summation, obtains hair color.

Determining a kind of feasible mode of the pupil color of personage's picture to be processed is: obtaining pupil region, is based on lesser ring of Merkel Domain obtains pupil color.That is, all pixels point progress weighted sum obtains pupil face in the pupil region that will acquire Color can also be wherein the source for obtaining pupil region, which can be from personage's picture to be processed, extracts pupil region picture from eye First picture of eyeball extracts pupil region picture.

In this embodiment, it determines that the feasible mode of pupil region has but is not limited to: being recognized in personage's picture to be processed Key point of the pupil in eyes can determine pupil region based on these key points, the region formed such as these key points It can be considered as pupil region.

S106, second picture, the skin color of personage's picture to be processed, figure map to be processed based on each target site The pupil color of the hair color of piece and personage's picture to be processed obtains the personage's picture shown with special image.It is understood that Are as follows: the skin color of the second picture of obtained each target site and personage's picture to be processed is synthesized into complete figure map Piece, at least one of nose, mouth, eyes, ear, eyebrow and hair in personage's picture target site is with special form As displaying, which specific target site shown with special image need by extracting the first picture of which target site and It is fixed.

In the present embodiment, a kind of feasible pattern for obtaining the personage's picture shown with special image is: determining each mesh Mark the second picture position and corresponding direction at position；Second picture position and institute based on each target site are right Direction is answered, the second picture of each target site is spliced, obtains the personage's picture shown with special image；It will be with special The skin color of personage's picture of image display is set as the skin color of personage's picture to be processed.

The process of the second picture position and corresponding direction that wherein determine each target site may is that with each Position and direction of the target site in personage's picture to be processed, determine position of the second picture of each target site in face It sets and direction can be according to people to be processed with the position where the second picture of the nose of special image displaying by taking nose as an example Nose position is foundation in object picture, determines the position of the second picture, such as determines the left and right wing of nose apart from left and right face Distance and bridge of the nose the top (be place that the bridge of the nose most start) distance apart from forehead of the distance, nose of cheek apart from chin, Position of the second picture of nose in face can be then determined with these.And the second figure of the nose shown with special image Direction corresponding to piece then depending on the face of personage's picture to be processed be positive face shine and side face shine depending on, as face be positive face shine when, with Before the direction of the second picture for the nose that special image is shown is positive, when face is left side of the face, with the nose of special image displaying The direction of the second picture of son is left side.

After the second picture position and corresponding direction for obtaining each target site, in each target site Any position for, placed based on its position and corresponding direction, then spelled after the placement of any target site in this way It connects to obtain the personage's picture shown with special image, and then sets personage to be processed for skin color obtained in step S105 The skin color of picture sets obtained hair color to the hair color of personage's picture to be processed, the pupil face that will be obtained Color is set as the pupil color of personage's picture to be processed, i.e., by shown with special image each target site second picture, to Handle skin color, the hair color of personage's picture to be processed and the pupil color knot of personage's picture to be processed of personage's picture It closes, obtains the personage's picture of better authenticity shown with special image.

From above-mentioned technical proposal it is found that being mentioned from personage's picture to be processed in the case where obtaining personage's picture to be processed The first picture for taking out each target site, to the either objective position in each target site: by the first of the target site Picture is determined as the input of the corresponding default identification model of the target site, obtains the corresponding default identification model of the target site The mark of the target site of output, the mark based on the target site obtain the of the target site from default picture library Two pictures, wherein the second picture of the target site shows the target site with special image；And determining figure map to be processed In the case where the skin color of piece, the skin color of second picture and personage's picture to be processed based on each target site is obtained To the personage's picture shown with special image, personage's picture to be processed is automatically converted to special image displaying to realize Personage's picture, and above-mentioned transform mode can choose second picture from default picture library automatically, so that user's need not have Fine arts grounding in basic skills can also obtain the special image for meeting user's esthetic requirement, and above-mentioned conversion is not necessarily to user from default picture library Interested second picture is chosen, manually so as to reduce the complexity of conversion.

For below using user's self-timer as personage's picture to be processed, picture provided in this embodiment is handled in conjunction with Fig. 3 Method is illustrated, in Fig. 3 using personage's picture to be processed as input, by picture pretreatment (such as above-mentioned recognition of face or Picture extracts model) the first picture of multiple target sites is obtained, eyes picture and nose picture are only shown, but practical in Fig. 3 It can obtain the first picture of eyes, nose, mouth, eyebrow, ear and hair these target sites, and then by these target sites The first picture be input in corresponding default identification model and (eyes picture be input to eye model in such as Fig. 3), by (eyes number as shown in Figure 3, i.e., eyes are in default picture library for corresponding default respective mark of identification model output Number), it is selected from default picture library based on respective mark and shows the second picture of target site (as schemed with cartoon character Eyes part shown in 3, i.e. second picture), finally by the splicing for the second picture for showing these target sites with cartoon character, The personage's picture shown with special image is obtained, and the skin color of personage's picture is the skin face of personage's picture to be processed Color, to keep the consistent of skin color.

Referring to Fig. 4, it is illustrated, the embodiment of the invention provides the flow charts of another image processing method, for leading to User feedback is crossed to correct default identification model, to improve users satisfaction degree, which may include following step It is rapid:

S201 obtains personage's picture to be processed.

S202, extracts the first picture of each target site from personage's picture to be processed, and each target site includes: At least one position in nose, mouth, eyes, ear, eyebrow and hair.

First picture of the target site is determined as the input of the corresponding default identification model of the target site by S203, Obtain the mark of the target site of the corresponding default identification model output of the target site.

S204, to the either objective position in each target site: the mark based on the target site, from default picture library The middle second picture for obtaining the target site, wherein the second picture of the target site shows the target site with special image.

S205 determines the skin color of personage's picture to be processed, the hair color of personage's picture to be processed and people to be processed The pupil color of object picture.

S206, the skin color of second picture and personage's picture to be processed based on each target site, personage to be processed The pupil color of the hair color of picture and personage's picture to be processed obtains the personage's picture shown with special image.

In the present embodiment, above-mentioned S201 to S206 is identical as the implementation procedure of above-mentioned S101 to S106 and principle, this In repeat no more.

S207 is obtained field feedback, and is modified based on field feedback to default identification model.Specifically : obtain user to the feedback information of personage's picture of feature image display, feedback information be at least used to show user to The satisfaction for the face picture that special image is shown is based on feedback information, is corrected again to default identification model, to repair Change the corresponding relationship between the mark of the target site of input and the output of default identification model of default identification model.

Obtain user be to a kind of feasible pattern of the feedback information of the personage's picture shown with special image: display with After personage's picture that special image is shown, display one, for investigating the table of satisfaction, may include: full in the table Meaning degree and modification mode etc., wherein satisfaction shows whether the personage's picture shown with special image meets user to special The requirement of image, modification mode then demonstrate the need for modifying to which position of the personage's picture shown with special image, Its desired second picture can also be further chosen from default picture library by user, these information can be carried anti- In feedforward information.The input and default identification of default identification model are modified by modified mode again after obtaining feedback information Corresponding relationship between the mark of the target site of model output.

Such as the corresponding relationship at the modification position in default identification model is carried out based on the modification position that feedback information carries Modification, or the corresponding relationship at the modification position in default identification model is revised as the second figure that user in feedback information specifies Number of the piece in default picture library improves user so that the second picture of subsequent output is the second picture that user wants Satisfaction, wherein modification position is the determining needs of any user in above-mentioned nose, mouth, eyes, ear, eyebrow and hair The position of modification.

From above-mentioned technical proposal it is found that correcting automatically by feedback information to pre-identification model, default identification can be modified Corresponding relationship between the mark of the target site of the input of model and the output of default identification model, so that the standard of pre-identification model True rate is higher and higher and increasingly meets user's expectation, so that the personage's picture shown with special image is close to true face, So as to improve users satisfaction degree.

For the various method embodiments described above, for simple description, therefore, it is stated as a series of action combinations, but Be those skilled in the art should understand that, the present invention is not limited by the sequence of acts described because according to the present invention, certain A little steps can be performed in other orders or simultaneously.Secondly, those skilled in the art should also know that, it is retouched in specification The embodiment stated belongs to preferred embodiment, and related actions and modules are not necessarily necessary for the present invention.

Corresponding with above method embodiment, the embodiment of the present invention provides a kind of picture processing unit, structural schematic diagram As shown in figure 5, may include: that obtain module 11, extraction module 12, identification module 13, second picture determining module 14, color true Cover half block 15 and synthesis module 16.

Module 11 is obtained, for obtaining personage's picture to be processed, in the present embodiment, module 11 is obtained and obtains tape handling people The mode of object picture can be, by the picture of user's upload, the picture received from other equipment and under in other equipment At least one of picture of load.

Extraction module 12, for extracting the first picture of each target site, each mesh from personage's picture to be processed Mark position includes: at least one position in nose, mouth, eyes, ear, eyebrow and hair, that is to say, that from people to be processed One or more target sites are extracted in object picture, such as personage's picture to be processed can be the figure including target complete position Piece, be also possible to include partial target position picture, then the first picture of each target site extracted and personage to be processed The target site that picture includes is related.

In the present embodiment, extraction module 12 extracts the first picture of each target site from personage's picture to be processed A kind of feasible pattern be: to figure map's piece to be processed carry out recognition of face, by identification obtain nose, mouth, eyes, ear, At least one position in eyebrow and hair, by least one position institute in nose, mouth, eyes, ear, eyebrow and hair Cut in region, target site separated from personage's picture to be processed, thus obtain nose, mouth, eyes, First picture at the either objective position of ear, eyebrow and hair, such as above-mentioned each target site is obtained by recognition of face Multiple key points, to either objective position in each target site: determining the mesh by multiple key points of the target site Position region is marked, then extracts in determined region from personage's picture to be processed, obtains the of the target site One picture.

Another feasible pattern that extraction module 12 extracts the first picture of each target site is: training obtains figure in advance Piece extracts model, and personage's picture to be processed is input to the picture and is extracted in model, obtains the picture and extracts each of model output First picture of a target site can also train in advance in practical applications certainly and obtain multiple pictures extraction models, each Picture extracts the corresponding target site of model, i.e. a picture extracts the first figure that model is used to extract a target site Piece extracts the process of model and may is that through cross validation mode from default personage's pictures wherein training in advance obtains picture Training set and test set are determined in conjunction, to any personage's picture in training set: marking out from personage's picture Each target site, each target site based on mark carry out model training, obtain picture and extract model, to be mentioned by picture Modulus type obtains the key point of each target site, and extracts model by picture and exported based on the key point of each target site Then first picture of each target site extracts model to obtained picture by each personage's picture in test set again It is tested, determines that picture extracts model and can extract the first picture of each target site if testing, terminate to train, otherwise Again personage's picture in training set is labeled and is trained again.

Additionally need to illustrate a bit: extraction module 12, which extracts model from picture, can be directed at least one people, that is, It says that a picture extracts model and can be only applicable to a people, picture is obtained by personage's picture training of a people and extracts mould Type or a picture extract model and can be adapted for multiple people, obtain picture by personage's picture training of multiple people and extract Model.

Identification module 13, for the either objective position in each target site: by the first picture of the target site As the input of the corresponding default identification model of the target site, the corresponding default identification model output of the target site is obtained The mark of the target site, the mesh that wherein the corresponding default identification model of the target site will be extracted from historical personage's picture First picture at position is marked as input, is trained to obtain using the mark marked for the target site as output.

In the present embodiment, identification module 13 obtains the process of default identification model are as follows: obtains historical personage's figure of face The set of piece marks off training set and test set, by recognition of face or above-mentioned from the set of historical personage's picture Picture extracts model and obtains the first picture of a target site in any historical personage's picture in training set, and marks out one The mark of a target site carries out model training with the first picture of the target site and the mark marked out, obtains corresponding Default identification model, wherein the default identification model that training obtains is corresponding with target site respectively, that is to say, that each mesh There is corresponding default identification model at mark position, as nose target site has corresponding default identification model, the nose got First picture at sub-goal position will be input to the corresponding default identification model of nose target site；Then pass through test set Each historical personage's picture in conjunction tests default identification model, to determine whether default identification model meets the requirements, The optimization to default identification model can be realized by repetition training and test in the case where undesirable, set for what is required This present embodiment can no longer be illustrated depending on practical application by setting.

It in practical annotation process, needs to classify to either objective position, and is directed to for either objective position, it is right The special image of face and the shape of user in the thinner personage's picture that can to show with special image of the classification of the target site As being more bonded, such as by taking the hair of women as an example, the hair of women can be divided into but be not limited to long hair, middle hair, bob and super Bob can be divided into big wave curly hair, bubble roll hair and straight hair in long hair again, then need to distinguish for these types in assorting process A mark is marked out, it is consistent or close with the feature having by the corresponding type of the same mark, but different identification The distinctiveness for the feature that corresponding type has becomes apparent from, and can make same text in datagram valut by the classification in this way (the first picture in such as same file has the similarity between the first picture in part folder in default similarity dimensions Feature it is consistent or close), the higher the better for the degree of isolation of the first picture of different files, still by taking the hair of women as an example, The hair style of the hair of all first pictures in No. 00 file is all consistent as far as possible, and the first figure in No. 01 file Hair style in piece will be with the difference of the hair style in the first picture in No. 00 file as far as possible.

Wherein No. 00 and No. 01 can then be considered as mark of this target site of hair in data picture library, have same The default similarity of all first pictures of one mark in default similarity dimensions, all first pictures of different identification it is pre- If similarity should be less than default similarity dimensions, meanwhile, the quantity of the first picture with like-identified is greater than or equal to 1, Being exactly includes at least one secondary first picture in certain mark, includes at least one first in file such as by taking file as an example Picture, and the similarity between each the first picture in same file folder is being preset in similarity dimensions, different files The default similarity between each the first picture in folder is not in default similarity dimensions.In the present embodiment, identification module 13 use SimpleNet picture classification mode as the classification to target site, are specifically chosen SimpleNet reason in a kind of figure It has been illustrated in piece processing method, has no longer repeated herein.

Second picture determining module 14, for the either objective position in each target site: being based on the target site Mark, the second picture of the target site is obtained from default picture library, wherein the second picture of the target site is with special The image display target site.

Target site is identified as presetting the number in picture library by target site, second picture determining module 14 A kind of feasible pattern from the second picture for obtaining the target site in default picture library is: determining the mesh in default picture library Number corresponding each width second picture of the position in default picture library is marked, chooses a width from identified each width second picture Picture, specifically the same number corresponds to a file in default picture library, can have one or more under a file Second picture obtains a file according to number of each target site in default picture library, and then from this document folder It determines second picture of the width second picture as the target site, can such as randomly select a width or according to the preferential of picture Grade is chosen, and wherein the priority of picture can be depending on the number that picture is selected, and specific the present embodiment no longer illustrates.With nose For, the number based on nose in default picture library is obtained with special image from default picture library, as cartoon character is shown The second picture of nose.

Color determination module 15, for determining the skin color of personage's picture to be processed, the hair of personage's picture to be processed The pupil color of color and personage's picture to be processed.In the present embodiment, color determination module 15 determines skin color, hair face The mode of color and pupil color is following but is not limited to such as under type:

Determining a kind of feasible mode of the skin color of personage's picture to be processed is: determining the skin in personage's picture to be processed Skin region, wherein skin area be personage's picture to be processed face and/or face's following region in meet pre-set color condition Region；Interference of the filtering environmental factor to the skin color in skin area obtains the skin color of personage's picture to be processed.

Determining a kind of feasible mode of the pupil color of personage's picture to be processed is: obtaining pupil region, is based on lesser ring of Merkel Domain obtains pupil color.That is, all pixels point progress weighted sum obtains pupil face in the pupil region that will acquire Color, wherein the source for obtaining pupil region, which may is that from processing personage's picture, extracts pupil region picture, it may also is that from eye The first picture of eyeball extracts pupil region picture.

Above-mentioned determining skin color, hair color and the mode of pupil color are please referred to related in embodiment of the method Illustrate, this present embodiment is not being illustrated.

Synthesis module 16, for second picture, personage's picture to be processed based on each target site skin color, to The hair color of personage's picture and the pupil color of personage's picture to be processed are handled, the figure map shown with special image is obtained Piece.

In the present embodiment, a kind of structure of synthesis module 16 is as shown in fig. 6, synthesis module 16 may include: to determine list Member 161, concatenation unit 162 and synthesis unit 163, wherein determination unit 161 is used to determine the second picture of each target site Position and corresponding direction；Concatenation unit 162 for based on each target site second picture position and institute it is right Direction is answered, the second picture of each target site is spliced, obtains the personage's picture shown with special image；Synthesis unit 163 for setting the skin color of the personage's picture shown with special image to the skin color of personage's picture to be processed.

Such as the second picture position of the determining each target site of determination unit 161 and the process in corresponding direction can To be: with position and direction of each target site in personage's picture to be processed, determining the second picture of each target site Position and direction in face, by taking nose as an example, the position where the second picture of the nose shown with special image can be with It is foundation according to nose position in personage's picture to be processed, determines the position of the second picture, such as determines left and right nose Distance and the bridge of the nose the top (being the place that the bridge of the nose most starts) distance that foilbase is with a distance from the cheek of left and right, nose is apart from chin The distance of forehead can then determine position of the second picture of nose in face with these.And shown with special image Direction corresponding to the second picture of nose then depending on the face of personage's picture to be processed be positive face shine and side face shine depending on, as face is When positive face shines, before the direction of the second picture of the nose shown with special image is positive, when face is left side of the face, with special form As the direction of the second picture of the nose of displaying is left side.

After concatenation unit 162 obtains the second picture position and corresponding direction of each target site, to each For any position in target site, is placed based on its position and corresponding direction, put in this way in any target site Then splicing obtains the personage's picture shown with special image after setting, and then is set obtained skin color by synthesis unit 163 It is set to the skin color of personage's picture to be processed, sets obtained hair color to the hair color of personage's picture to be processed, It sets obtained pupil color to the pupil color of personage's picture to be processed, i.e., will show each target site with special image Second picture, the skin color of personage's picture to be processed, the hair color of personage's picture to be processed and personage's picture to be processed Pupil color combine, obtain better authenticity with special image show personage's picture.

From above-mentioned technical proposal it is found that being mentioned from personage's picture to be processed in the case where obtaining personage's picture to be processed First picture at the either objective position in each target site is determined as presetting by the first picture for taking out each target site The input of identification model obtains the mark of the target site of default identification model output, to any in each target site Target site: the mark based on the target site obtains the second picture of the target site from default picture library, wherein the mesh The second picture at mark position shows the target site with special image；And in the feelings for the skin color for determining personage's picture to be processed Under condition, the skin color of second picture and personage's picture to be processed based on each target site obtains showing with special image Personage's picture, thus realize by personage's picture to be processed be automatically converted to special image show personage's picture, and on Second picture can be chosen from default picture library automatically by stating transform mode, so that user's need not have fine arts grounding in basic skills can also obtain To the special image for meeting user's esthetic requirement, and above-mentioned conversion chosen manually from default picture library without user it is interested Second picture, so as to reduce the complexity of conversion.

The embodiment of the present invention also provides another picture processing unit, and structural schematic diagram is as shown in fig. 7, may include: Obtain module 21, extraction module 22, identification module 23, second picture determining module 24, color determination module 25, synthesis module 26 With correction module 27.

Module 21 is obtained, for obtaining personage's picture to be processed.

Extraction module 22 extracts the first picture of each target site, each target portion from personage's picture to be processed Position includes: at least one position in nose, mouth, eyes, ear, eyebrow and hair.

Identification module 23, for the either objective position in each target site: by the first picture of the target site It is determined as the input of the corresponding default identification model of the target site, obtains the corresponding default identification model output of the target site The target site mark.

Second picture determining module 24, for the either objective position in each target site: being based on the target site Mark, the second picture of the target site is obtained from default picture library, wherein the second picture of the target site is with special The image display target site.

Color determination module 25, for determining the skin color of personage's picture to be processed, the hair of personage's picture to be processed The pupil color of color and personage's picture to be processed.

Synthesis module 26, for based on each target site second picture and personage's picture to be processed skin color, The pupil color of the hair color of personage's picture to be processed and personage's picture to be processed obtains the figure map shown with special image Piece.

In the present embodiment, above-mentioned acquisition module 21, extraction module 22, identification module 23, second picture determining module 24, Color determination module 25, synthesis module 26 and above-mentioned acquisition module 11, extraction module 12, identification module 13, second picture determine Module 14, color determination module 15, the implementation procedure of synthesis module 16 and principle are identical, and which is not described herein again.

Correction module 27 carries out default identification model for obtaining field feedback, and based on field feedback Amendment.Correction module 27 includes acquiring unit 271 and amending unit 272, and specific structure please join Fig. 8.Acquiring unit 271 is for obtaining User is obtained to the feedback information of personage's picture of feature image display；Feedback information is at least used to show user to special form As the satisfaction of the face picture of displaying, amending unit 272 is used to be based on feedback information, carries out again to default identification model Amendment, the input pass corresponding between the mark for the target site that default identification model exports to modify default identification model System.

Wherein acquiring unit 271 is for obtaining user to one kind of the feedback information of the personage's picture shown with special image Feasible pattern is: after personage's picture that display is shown with special image, display one is used to investigate the table of satisfaction, May include in the table: satisfaction and modification mode etc., wherein satisfaction shows the figure map shown with special image Whether piece meets requirement of the user to special image, and modification mode is then demonstrated the need for the personage's picture shown with special image Which position modify, its desired second picture can also be further chosen from default picture library by user, this A little information can carry in feedback information.Default identification is modified by modified mode again after obtaining feedback information Corresponding relationship between the mark of the target site of the input of model and the output of default identification model.

Amending unit 272 is corrected a kind of feasible pattern to default identification model again: being carried based on feedback information Modification position modify to the corresponding relationship at the modification position in default identification model, or will in default identification model should The corresponding relationship at modification position, which is revised as the second picture that user in feedback information specifies, is presetting the number in picture library, so that The second picture that the second picture of subsequent output is wanted for user is obtained, users satisfaction degree is improved, wherein modification position is above-mentioned Any user in nose, mouth, eyes, ear, eyebrow and hair determines the position for needing to modify.

It should be noted that all the embodiments in this specification are described in a progressive manner, each embodiment weight Point explanation is the difference from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For device class embodiment, since it is basically similar to the method embodiment, so being described relatively simple, related place ginseng See the part explanation of embodiment of the method.

Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged Except there is also other identical elements in the process, method, article or apparatus that includes the element.

The foregoing description of the disclosed embodiments can be realized those skilled in the art or using the present invention.To this A variety of modifications of a little embodiments will be apparent for a person skilled in the art, and the general principles defined herein can Without departing from the spirit or scope of the present invention, to realize in other embodiments.Therefore, the present invention will not be limited It is formed on the embodiments shown herein, and is to fit to consistent with the principles and novel features disclosed in this article widest Range.

The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered It is considered as protection scope of the present invention.

Claims

1. a kind of image processing method, which is characterized in that the described method includes:

Obtain personage's picture to be processed；

The first picture of each target site is extracted from personage's picture to be processed, each target site includes: At least one position in nose, mouth, eyes, ear, eyebrow and hair；

To the either objective position in each target site: the first picture of the target site is determined as the target site The input of corresponding default identification model obtains the mark of the target site of the corresponding default identification model output of the target site Know, wherein the corresponding default identification model of the target site is by the first figure of the target site extracted from historical personage's picture Piece is trained to obtain as input using the mark marked for the target site as output；

To the either objective position in each target site: the mark based on the target site is somebody's turn to do from default picture library The second picture of target site, wherein the second picture of the target site shows the target site with special image；

Determine the skin color of personage's picture to be processed, the hair color of personage's picture to be processed and described to be processed The pupil color of personage's picture；

Second picture, the skin color of personage's picture to be processed, the figure map to be processed based on each target site The pupil color of the hair color of piece and personage's picture to be processed obtains the personage's picture shown with special image.

2. the method according to claim 1, wherein the skin color of determination personage's picture to be processed Include:

Determine that the skin area in personage's picture to be processed, the skin area are the faces of personage's picture to be processed And/or meet the region of pre-set color condition in face's following region；

Interference of the filtering environmental factor to the skin color in the skin area obtains the skin of personage's picture to be processed Color.

3. the method according to claim 1, wherein the default identification model exports the target site in institute The number in default picture library is stated, number of the target site in the default picture library is determined as the target site Mark, and at least width second picture for including in the corresponding default picture library of the same number, an at least width The similarity between each width picture in second picture is in default similarity dimensions；

The either objective position in each target site: the mark based on the target site is obtained from default picture library The second picture for obtaining the target site includes: to determine the target site in the default picture library in the default picture library The corresponding each width second picture of number, choose a width picture from identified each width second picture.

4. the method according to claim 1, wherein the second picture based on each target site, described The pupil of the skin color of personage's picture to be processed, the hair color of personage's picture to be processed and personage's picture to be processed Hole color obtains the personage's picture shown with special image:

Second picture position and corresponding direction based on each target site, to the second picture of each target site into Row splicing, obtains the personage's picture shown with special image；

The people to be processed is set by skin color, hair color and the pupil color of the personage's picture shown with special image Skin color, hair color and the pupil color of object picture.

5. the method according to claim 1, wherein the method also includes:

User is obtained to the feedback information of personage's picture with feature image display, the feedback information is at least used to show Satisfaction of the user to the face picture shown with special image；

Based on the feedback information, the default identification model is corrected again, to modify the default identification model Corresponding relationship between the mark of the target site of input and the default identification model output.

6. a kind of picture processing unit, which is characterized in that described device includes:

Module is obtained, for obtaining personage's picture to be processed；

Extraction module, it is described each for extracting the first picture of each target site from personage's picture to be processed Target site includes: at least one position in nose, mouth, eyes, ear, eyebrow and hair；

Identification module, for the either objective position in each target site: the first picture of the target site is true It is set to the input of the corresponding default identification model of the target site, obtains the corresponding default identification model output of the target site The mark of the target site, the mesh that wherein the corresponding default identification model of the target site will be extracted from historical personage's picture First picture at position is marked as input, is trained to obtain using the mark marked for the target site as output；

Second picture determining module, for the either objective position in each target site: the mark based on the target site, The second picture of the target site is obtained from default picture library, wherein the second picture of the target site is shown with special image The target site；

Color determination module, for determining the skin color of personage's picture to be processed, the head of personage's picture to be processed Send out the pupil color of color and personage's picture to be processed；

Synthesis module, the skin color, described for second picture, personage's picture to be processed based on each target site The pupil color of the hair color of personage's picture to be processed and personage's picture to be processed obtains the people shown with special image Object picture.

7. device according to claim 6, which is characterized in that the color determination module, be specifically used for determining it is described to The skin area in personage's picture, and interference of the filtering environmental factor to the skin color in the skin area are handled, is obtained The skin color of personage's picture to be processed, the skin area are the face and/or face of personage's picture to be processed Meet the region of pre-set color condition in following region.

8. device according to claim 6, which is characterized in that the identification module exists for exporting the target site Number of the target site in the default picture library is determined as the target portion by the number in the default picture library The mark of position, and at least width second picture for including in the corresponding default picture library of the same number, described at least one The similarity between each width picture in width second picture is in default similarity dimensions；

The second picture determining module, for determining the target site in the default picture library in the default picture library The corresponding each width second picture of number, choose a width picture from identified each width second picture.

9. device according to claim 6, which is characterized in that the synthesis module comprise determining that unit, concatenation unit and Synthesis unit；

The concatenation unit, for second picture position and corresponding direction based on each target site, to each mesh The second picture at mark position is spliced, and the personage's picture shown with special image is obtained；

The synthesis unit, skin color, hair color and the pupil color of personage's picture for will be shown with special image It is set as skin color, hair color and the pupil color of personage's picture to be processed.

10. the device as claimed in claim 6, which is characterized in that described device further includes correction module, the correction module It include: obtaining unit and amending unit；

The acquiring unit, it is described anti-for obtaining user to the feedback information of personage's picture with feature image display Feedforward information is at least used to show the user to the satisfaction of the face picture shown with special image；

The amending unit is corrected the default identification model, again for being based on the feedback information to modify State the corresponding relationship between the mark of the target site of input and the default identification model output of default identification model.