CN109389019A

CN109389019A - Facial image selection method, device and computer equipment

Info

Publication number: CN109389019A
Application number: CN201710692300.4A
Authority: CN
Inventors: 何海峰; 钮毅
Original assignee: Hangzhou Hikvision Digital Technology Co Ltd
Current assignee: Hangzhou Hikvision Digital Technology Co Ltd
Priority date: 2017-08-14
Filing date: 2017-08-14
Publication date: 2019-02-26
Anticipated expiration: 2037-08-14
Also published as: CN109389019B

Abstract

The embodiment of the invention provides facial image selection method, device and computer equipment, facial image selection method includes: to obtain facial image group；Based on the first kind face character value of each facial image in facial image group, the goal gradient of each facial image is determined；Construct the corresponding multiple alternative image groups of facial image group；After each alternative image group is constituted, based on the goal gradient of each facial image in the alternative image group, the screening rule for being directed to the alternative image group is determined, and pass through corresponding screening rule, facial image in the alternative image group is screened, facial image to be utilized is obtained；Based on all facial images to be utilized corresponding to facial image group, target facial image corresponding to facial image group is determined.Facial image selection method is provided using the embodiment of the present invention, the accuracy rate of recognition of face can be improved.

Description

Facial image selection method, device and computer equipment

Technical field

The present invention relates to technical field of image processing, more particularly to a kind of facial image selection method, device and calculating Machine equipment.

Background technique

Recognition of face, usually also referred to as Identification of Images, face recognition, are the active research directions in area of pattern recognition One of, it is a kind of biological identification technology for carrying out identification based on facial feature information of people.The technology is set using camera shooting Standby image of the acquisition containing face, and automatic detection and tracking face in the acquired images, and then to the face detected It is identified.

The detailed process of above-mentioned recognition of face are as follows: target facial image is determined from facial image group, wherein facial image Group is made of the collected facial image for being directed to a people, and facial image is the people captured by Face datection algorithm The image in face region, target facial image are the higher facial images of mass ratio；Calculate target facial image and pre-recorded The similarity of the corresponding template facial image of target facial image；Based on the similarity being calculated, corresponding identification knot is exported Fruit, for example, similarity is not less than pre-set threshold value, then recognition result is that successfully, otherwise, recognition result is failure.

During above-mentioned recognition of face, can the target facial image of selection to identifying successfully there is very big shadow It rings.Currently, the method for facial image selection are as follows: for each facial image in facial image group, calculate the facial image Clarity, face size and human eye opening degree these three face characters；For each facial image, based on for the face figure As clarity, face size and the human eye opening degree being calculated, the overall merit score of the facial image is determined；By highest The corresponding facial image of overall merit score, be determined as target facial image.

In above-mentioned facial image selection method, overall merit score is clarity, face size and human eye opening degree The arithmetic sum of the attribute value of these three face characters.In this case, the corresponding face figure of selection highest overall merit score As being used as target facial image, in three face characters of target facial image may one of face character attribute value very It is small, but the attribute value of other two face characters is bigger.And if the face character of attribute value very little is to the matter of facial image Amount influence is bigger, then will lead to the poor quality of target facial image.And then due to the matter of selected target facial image Amount is not very well, to cause the probability of misrecognition of the target facial image identification face using selection higher, reduce recognition of face Accuracy rate.

Summary of the invention

The embodiment of the present invention is designed to provide a kind of facial image selection method, device, computer equipment and calculating Machine readable storage medium storing program for executing, to improve the accuracy rate of recognition of face.Specific technical solution is as follows:

In a first aspect, in order to achieve the above objectives, the embodiment of the invention provides a kind of facial image selection method, the sides Method includes:

Obtain facial image group, wherein include multiple facial images in the facial image group；

Based on the first kind face character value of each facial image in the facial image group, each facial image is determined Goal gradient, wherein the first kind face character value is the corresponding value of first kind face character, the first kind face character Include at least one face character；

Construct the corresponding multiple alternative image groups of the facial image group, wherein the multiple alternative image group covers institute The face images in facial image group are stated, and each alternative image group includes at least two facial images；

After each alternative image group is constituted, based on the goal gradient of each facial image in the alternative image group, determine For the screening rule of the alternative image group, and by corresponding screening rule, to the facial image in the alternative image group It is screened, obtains facial image to be utilized；

Based on all facial images to be utilized corresponding to the facial image group, determine corresponding to the facial image group Target facial image.

Optionally, the step of building facial image group corresponding multiple alternative image groups, comprising:

The facial image of the first preset quantity is selected from the facial image group, constructs first alternative image group；

Alternative for any non-first image group determines the facial image group currently non-selected facial image Destination number, if the destination number is greater than the second preset quantity, from the current non-selected face of the facial image group Select the facial image of the second preset quantity in image, and using the facial image of selected second preset quantity and from upper one The facial image to be utilized that a alternative image group is screened constructs an alternative image group；If the destination number is little In the second preset quantity, the current non-selected face images and from a upper alternative image of the facial image group are utilized The facial image to be utilized that group screening obtains, constructs an alternative image group.

Facial image in the facial image group multiple alternative image groups are not divided into；

Alternatively,

Facial image in the facial image group is divided into multiple alternative image groups.

Optionally, the first kind face character value based on each facial image in the facial image group determines every The step of goal gradient of one facial image, comprising:

As follows, the goal gradient of each facial image in the facial image group is determined:

Determine the first face attribute value of current facial image；The first face attribute value is the first face character pair The value answered, first face character are a face character in the first kind face character；

Judge whether the first face attribute value meets the corresponding preset condition of first face character, In, first face character corresponds at least one preset condition, and a default item corresponding to first face character Part is the attribute-value ranges for the grade setting of the first face character, and a preset condition is corresponding with a grade；

When the judgment result is yes, by the corresponding grade of the preset condition met, it is determined as current facial image Goal gradient；

When the judgment result is no, a unemployed face character is selected from the first kind face character, it will First face character replaces with selected face character, returns to the first face category for executing and determining current facial image The step of property value.

Optionally, the first kind face character includes positive and negative face, clarity, brightness, coverage extent, deflection angle and pitching Angle.

Optionally, each face character in the first kind face character has correspondence at least one grade；

The goal gradient based on each facial image in the alternative image group determines the sieve for being directed to the alternative image group Select the step of rule, comprising:

Judge whether the facial image in the alternative image group meets the first default screening conditions, wherein described first is pre- If screening conditions are as follows: the goal gradient of the facial image in the alternative image group is all different, and is not simultaneously the second face category The corresponding grade of property；Second face character is a face character in the first kind face character；

If it is satisfied, pre-set first screening rule is determined as the screening rule for the alternative image group, In, first screening rule is the rule for screening the highest facial image of goal gradient.

Optionally, in the case where judging that the facial image in the alternative image group is unsatisfactory for the first default screening conditions, The method also includes:

Determine the third face attribute value of each facial image in the alternative image group；Wherein, the third face character Value be the corresponding value of third face character, the third face character be except the face character in the first kind face character it Outer face character；

Judge whether the facial image in the alternative image group meets the second default screening conditions, wherein described second is pre- If screening conditions are as follows: the goal gradient of the facial image in the alternative image group is that second face character is corresponding etc. simultaneously Maximum difference in grade and the difference being directed between the third face attribute value of every two facial image is greater than the first default threshold Value and the quantity of third facial image are less than the second preset threshold；The third facial image is the third face character Value is less than the facial image of third predetermined threshold value；

If it is satisfied, pre-set second screening rule is determined as the screening rule for the alternative image group；Its In, second screening rule is the rule for screening the highest facial image of the first comprehensive score；

First comprehensive score is through the following steps that be calculated:

For each facial image in the alternative image group, obtain each in the second class face character of the facial image The reference value of face character, wherein the second class face character includes third face character, and the reference value of face character is base Determining in the corresponding value of the face character；

For each facial image in the alternative image group, according in advance for the first of third face character setting Weight combination, is weighted the reference value of each face character of the facial image, obtain the facial image first is comprehensive Close scoring.

Optionally, in the case where judging that the facial image in the alternative image group is unsatisfactory for the second default screening conditions, The method also includes:

Judge whether the facial image in the alternative image group meets third and preset screening conditions, wherein the third is pre- If screening conditions are as follows: the goal gradient of the facial image in the alternative image group is the corresponding grade of second face character, And the grade difference between every two goal gradient is all larger than the 4th preset threshold；

If so, first screening rule is determined as the screening rule for the alternative image group.

Optionally, judge the facial image in the alternative image group be unsatisfactory for third preset screening conditions in the case where, The method also includes:

By pre-set third screening rule, it is determined as the screening rule for the alternative image group；Wherein, described Three screening rules are the rule for screening the highest facial image of the second comprehensive score；

Second comprehensive score is through the following steps that be calculated:

Determine the third class face character value of each facial image in the alternative image group；Wherein, the third class face Attribute value is the corresponding value of third class face character, and the third class face character is except the first kind face character and the The face character except face character in two class face characters；

According to the relationship that preset goal gradient is combined with the second weight, each facial image in the alternative image group is determined The second weight combination；

For each facial image in the alternative image group, according to determine the second weight of the facial image combine and Each face character value in 4th class face character value obtains the second comprehensive score of the facial image, wherein the described 4th Class face character value is the corresponding value of the 4th class face character, and the 4th class face character includes the first kind face category Face character in property, the second class face character and the third class face character.

Optionally, each facial image in the alternative image group, according to determining the of the facial image Each face character value in the combination of two weights and the 4th class face character value is weighted, and obtains the of the facial image The step of one comprehensive score, comprising:

For each facial image in the alternative image group, according to each in pre-set 4th class face character Mapping relations between the corresponding value of face character and value, determine each in the 4th class face character of the facial image The value of face character, wherein each face character corresponds to the same value range；

It is combined according to the second weight for determining the facial image, each face character corresponding to the facial image takes Value is weighted, and obtains the second comprehensive score of the facial image.

Optionally, second face character is deflection angle；

The third face character is interpupillary distance；

The second class face character includes interpupillary distance, pitch angle, coverage extent and deflection angle；

The third class face character includes whether as yin-yang face, opens and closes eyes and open and shut up；

The 4th class face character include clarity, brightness, coverage extent, deflection angle, pitch angle, interpupillary distance, whether be Yin-yang face opens and closes eyes and opens and shuts up.

Second aspect, in order to achieve the above objectives, the embodiment of the invention also provides a kind of facial image selection devices, special Sign is that described device includes:

Module is obtained, for obtaining facial image group, wherein include multiple facial images in the facial image group；

First determining module, for the first kind face character value based on each facial image in the facial image group, Determining the goal gradient of each facial image, wherein the first kind face character value is the corresponding value of first kind face character, The first kind face character includes at least one face character；

Module is constructed, for constructing the corresponding multiple alternative image groups of the facial image group, wherein the multiple alternative Image group covers the face images in the facial image group, and each alternative image group includes at least two face figures Picture；

First obtains module, for being based on each face figure in the alternative image group after each alternative image group is constituted The goal gradient of picture determines the screening rule for being directed to the alternative image group, and by corresponding screening rule, to the alternative figure As the facial image in group is screened, facial image to be utilized is obtained；

Second determining module, for determining institute based on all facial images to be utilized corresponding to the facial image group State target facial image corresponding to facial image group.

Optionally, the building module, comprising:

First building submodule is constructed for selecting the facial image of the first preset quantity from the facial image group First alternative image group；

Second building submodule, is used for image group alternative for any non-first, determines that the facial image group is current The destination number of non-selected facial image, if the destination number is greater than the second preset quantity, from the facial image The facial image of the second preset quantity is selected in the current non-selected facial image of group, and utilizes selected second present count The facial image of amount and the facial image to be utilized screened from upper one alternative image group construct an alternative image group； If the destination number is not more than the second preset quantity, the current non-selected all face figures of the facial image group are utilized Picture and the facial image to be utilized screened from upper one alternative image group construct an alternative image group.

Optionally, the building module, it is multiple specifically for the facial image in the facial image group not to be divided into Alternative image group；

Alternatively,

Optionally, described first module is obtained, for as follows, determining each face in the facial image group The goal gradient of image:

Described first obtains module, comprising:

Judging submodule, for judging whether the facial image in the alternative image group meets the first default screening conditions, Wherein, the described first default screening conditions are as follows: the goal gradient of the facial image in the alternative image group is all different, and different When be the corresponding grade of the second face character；Second face character is a face category in the first kind face character Property；

First determines submodule, for will set in advance in the case where the judging result of the judging submodule is to meet The first screening rule set is determined as the screening rule for the alternative image group, wherein first screening rule is screening The rule of the highest facial image of goal gradient.

Optionally, described device further include:

Third determining module is to determine that this is standby in ungratified situation for the judging result in the judging submodule Select the third face attribute value of each facial image in image group；Wherein, the third face attribute value is third face character Corresponding value, the third face character are the face character in addition to the face character in the first kind face character；

First judgment module, for judging whether the facial image in the alternative image group meets the second default screening item Part, wherein the second default screening conditions are as follows: the goal gradient of the facial image in the alternative image group is described the simultaneously Maximum in the corresponding grade of two face characters and the difference being directed between the third face attribute value of every two facial image is poor Value is greater than the quantity of the first preset threshold and third facial image less than the second preset threshold；The third facial image is The third face attribute value is less than the facial image of third predetermined threshold value；

4th determining module, for the judging result of first judging submodule be meet in the case where, will be preparatory The second screening rule being arranged is determined as the screening rule for the alternative image group；Wherein, second screening rule is sieve Select the rule of the highest facial image of the first comprehensive score；

Second obtains module, for obtaining the second of the facial image for each facial image in the alternative image group The reference value of each face character in class face character, wherein the second class face character includes third face character, people The reference value of face attribute is determining based on the corresponding value of the face character；

Third obtains module, for being directed to each facial image in the alternative image group, according in advance for the third The first weight combination of face character setting, is weighted the reference value of each face character of the facial image, obtains First comprehensive score of the facial image.

Optionally, described device further include:

Second judgment module is in ungratified situation for the judging result in the first judgment module, and judgement should Whether the facial image in alternative image group, which meets third, is preset screening conditions, wherein the third presets screening conditions are as follows: should The goal gradient of facial image in alternative image group is the corresponding grade of second face character, and every two goal gradient Between grade difference be all larger than the 4th preset threshold；

5th determining module, in the case where the judging result of the described the such as figure judgment module is to be, by described the One screening rule is determined as the screening rule for the alternative image group.

Optionally, described device further include:

6th determining module, for the judging result in second judgment module be ungratified situation under, will be preparatory The third screening rule of setting is determined as the screening rule for the alternative image group；Wherein, the third screening rule is sieve Select the rule of the highest facial image of the second comprehensive score；

7th determining module, for determining the third class face character value of each facial image in the alternative image group；Its In, the third class face character value is the corresponding value of third class face character, and the third class face character is except described the The face character except face character in a kind of face character and the second class face character；

8th determining module, the relationship for being combined according to preset goal gradient with the second weight determine the alternative figure As the second weight of facial image each in group combines；

4th obtains module, for for each facial image in the alternative image group, according to determining the face figure Each face character value in the second weight combination of picture and the 4th class face character value, obtain the facial image second are comprehensive Scoring, wherein the 4th class face character value is the corresponding value of the 4th class face character, and the 4th class face character includes Face character in the first kind face character, the second class face character and the third class face character.

Optionally, the described 4th module is obtained, comprising:

Second determines submodule, for for each facial image in the alternative image group, according to pre-set the The corresponding value of each face character in four class face characters and the mapping relations between value, determine the described of the facial image The value of each face character in 4th class face character, wherein each face character corresponds to the same value range；

Submodule is obtained, is combined according to the second weight for determining the facial image, it is corresponding to the facial image each The value of face character is weighted, and obtains the second comprehensive score of the facial image.

Optionally, second face character is deflection angle；

The third face character is interpupillary distance；

The third aspect, in order to achieve the above objectives, the embodiment of the invention also provides a kind of computer equipments, including processing Device, communication interface, memory and communication bus, wherein the processor, the communication interface, the memory pass through described Communication bus completes mutual communication；

The memory, for storing computer program；

The processor when for executing the program stored on the memory, realizes any of the above-described face figure As the method and step in selection method.

Fourth aspect, in order to achieve the above objectives, the embodiment of the invention also provides a kind of computer readable storage medium, institutes It states and is stored with computer program in computer readable storage medium, realized when the computer program is executed by processor State the method and step in any facial image selection method.

A kind of facial image selection method, device, computer equipment and readable storage medium storing program for executing provided in an embodiment of the present invention, The screening rule of alternative image group can be determined according to the goal gradient of the facial image in alternative image group, advised by screening It then determines facial image to be utilized, and then determines target image.Compared to the prior art, selection more has specific aim, needle To different alternative image groups, it can use different screening rules and screened, filter out the preferable facial image of mass ratio, And then improve the accuracy of recognition of face.Certainly, it implements any of the products of the present invention or method must be not necessarily required to reach simultaneously All the above advantage.

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 only this Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with It obtains other drawings based on these drawings.

Fig. 1 is a kind of flow diagram of facial image selection method provided in an embodiment of the present invention；

Fig. 2 is another flow diagram of facial image selection method provided in an embodiment of the present invention；

Fig. 3 is another flow diagram of facial image selection method provided in an embodiment of the present invention；

Fig. 4 is another flow diagram of facial image selection method provided in an embodiment of the present invention；

Fig. 5 is the structural schematic diagram of facial image selection device provided in an embodiment of the present invention；

Fig. 6 is the structural schematic diagram of computer equipment provided in an embodiment of the present invention.

Specific embodiment

Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other Embodiment shall fall within the protection scope of the present invention.

In order to improve the accuracy rate of recognition of face, the embodiment of the invention provides facial image selection method, device, calculating Machine equipment and computer readable storage medium.

A kind of facial image selection method provided in an embodiment of the present invention is introduced first below.

It should be noted that a kind of facial image selection method provided by the embodiment of the present invention is set applied to computer It is standby.In concrete application, which can be picture pick-up device, be also possible to non-picture pick-up device, wherein picture pick-up device includes Video camera, attendance record terminal or other intelligent terminals etc. with image collecting function, non-picture pick-up device are to adopt without image Collect the equipment of function.

Referring to Fig. 1, the embodiment of the invention provides a kind of facial image selection methods, include the following steps:

S101: facial image group is obtained, wherein include multiple facial images in the facial image group.Of the invention real It applies in example, the facial image in facial image group can be picture pick-up device acquisition, and picture pick-up device includes video camera, with image Acquire the intelligent terminal etc. of function.Wherein, picture pick-up device acquisition is when going out current moment from face to disappear the moment to face Between in section human face region image.

It is understood that when the computer equipment is picture pick-up device facial image group can be obtained from local；And work as When the computer equipment is non-picture pick-up device, the facial image group of other picture pick-up devices acquisition can be obtained, this is all reasonable.

S102: the first kind face character value based on each facial image in the facial image group determines each face The goal gradient of image, wherein the first kind face character value is the corresponding value of first kind face character, the first kind people Face attribute includes at least one face character.

It should be noted that first kind face character value is the corresponding value of face character in first kind face character.The The face character that a kind of face character is included is predetermined, specifically, can determines according to actual conditions, it can also root It is past empirically determined accordingly, it can also be determined according to influence degree of the face character to quality of human face image, be limited certainly and not only In this, herein without enumerating.

In embodiments of the present invention, after obtaining facial image group, the first kind attribute value of each facial image, needle are determined To a facial image, the first kind face character value based on the facial image determines the goal gradient of the facial image.

It is understood that being directed to different face characters, determine that the value method of face character is different.Illustratively, one A face character is brightness, another face character is clarity.Determine that the method for brightness value can be with are as follows: calculate the facial image Each pixel brightness value；Calculate the average value of the brightness value of all pixels point, the brightness value as the facial image.Really Determine the method for definition values are as follows: facial image is input to preparatory trained clarity classifier, obtains the facial image Definition values.Further, the clarity classifier can for for determine clarity neural network, support vector machine or Haar classifier etc., Haar classifier are the classifiers based on tree.Certainly, above-mentioned examples cited, the only present invention are real A specific example of example is applied, does not have restriction effect.

It should be noted that face character value can be a specific numerical value, it can be text, particularly due to face category Property itself property determine, it is, of course, also possible to by text conversion be number.Illustratively, face character is brightness, then the people Face attribute value is brightness value；Face character is to open to shut up, then face character value can be to open one's mouth or shut up, and face character value can also Think pre-set corresponding numerical value or the pre-set corresponding numerical value of shutting up of opening one's mouth.Certainly, above-mentioned examples cited, only One specific example of the embodiment of the present invention does not have restriction effect.

Determine that the mode of the goal gradient of each facial image can be with are as follows: the corresponding grade of one face character of selection is made For the goal gradient of the facial image；Alternatively, being transported to the face character value in the facial image first kind face character value It calculates, obtains operation result, the goal gradient by the corresponding grade of operation result, as the facial image.

The corresponding grade of one face character of selection mentioned here, as the goal gradient of the facial image, Ke Yiwei According to pre-set selection rule, a face character is selected；The corresponding grade of selected face character only one When, the goal gradient by the corresponding grade of the face character, as the facial image；It is corresponding etc. in selected face character When grade is multiple, a grade, the goal gradient as the facial image are selected.

S103: the corresponding multiple alternative image groups of the facial image group are constructed, wherein the multiple alternative image group is contained The face images in the facial image group are covered, and each alternative image group includes at least two facial images.

It is understood that multiple alternative image groups cover the face images in facial image group, can keep away in this way Exempt from the appearance for the case where facial image is missed.Each alternative image group includes at least two facial images, in this way can be rear In continuous process, according to the screening rule determined for alternative image group, the facial image in alternative image group is screened, So that the selection of facial image is more targetedly and detailed-oriented, so that it is determined that the facial image that mass is relatively good.If standby It selects and there was only a facial image in image group, then can not carry out subsequent screening.

Construct alternative image group, it can be understood as facial image is selected from facial image group, by selected face figure As the facial image in alternately image group.It should be noted that the facial image in different alternative image groups can be complete Portion is different, can also part it is identical.

The modes of the corresponding multiple alternative image groups of the facial image group is constructed there are a variety of, in a kind of specific implementation side In formula, the step of the building facial image group corresponding multiple alternative image groups, may include:

Alternatively,

It is understood that facial image not to be divided into multiple alternative image groups, then the people in all alternative image groups The quantity of face image is not all the same, it is possible to which the quantity of the facial image in the alternative image group in part is identical, it is also possible to all The quantity of facial image in alternative image group is all different.And the facial image in facial image group is divided into multiple alternative Image group illustrates that the quantity of the facial image in all alternative image groups is all the same.

S104: after each alternative image group is constituted, based on the goal gradient of each facial image in the alternative image group, The screening rule for being directed to the alternative image group is determined, and by corresponding screening rule, to the face in the alternative image group Image is screened, and facial image to be utilized is obtained.

In embodiments of the present invention, one can be selected from the alternative image group in the goal gradient of each facial image The corresponding screening rule of selected goal gradient is determined as the screening rule for the alternative image group by a goal gradient； The goal gradient of each facial image in the alternative image group can also be subjected to operation, according to operation result, determination is directed to The screening rule of the alternative image group；It can also judge the item that the goal gradient of the facial image of the alternative image group is met The corresponding screening rule of the condition met is determined as the screening rule for the alternative image group by part.

It should be noted that each screening rule is pre-set, different screening rule, to the sieve of facial image Select mode different.In addition, the facial image to be utilized screened from an alternative image group, it is alternative to can be used as another Facial image to be screened in image group.

S105: based on all facial images to be utilized corresponding to the facial image group, the facial image group is determined Corresponding target facial image.

It, can be from being needed it should be noted that an alternative image group at least filters out a facial image to be utilized Using one facial image of facial image group selection, as target facial image, certainly, target facial image is from needed benefit With one top-quality facial image of selection in facial image.

In embodiments of the present invention, according to the goal gradient of the facial image in alternative image group, alternative image group is determined Screening rule, facial image to be utilized is determined by screening rule, and then determine target image.Compared to the prior art, it selects Selecting more there is specific aim can use different screening rules for different alternative image groups and be screened, be filtered out The preferable facial image of mass ratio, and then improve the accuracy of recognition of face.

Below with reference to specific embodiment, it is provided for the embodiments of the invention a kind of facial image selection method and is situated between It continues.

As shown in Fig. 2, a kind of facial image selection method provided by the embodiment of the present invention, may include steps of:

S201: facial image group is obtained, wherein include multiple facial images in the facial image group.

S202: the first kind face character value based on each facial image in the facial image group determines each face The goal gradient of image, wherein the first kind face character value is the corresponding value of first kind face character, the first kind people Face attribute includes at least one face character.

It should be noted that S201-S202 is identical as the S101-S102 in above-described embodiment, this will not be repeated here.

S203: selecting the facial image of the first preset quantity from the facial image group, constructs first alternative image Group.

It should be noted that the first preset quantity can be the empirical value of artificial settings, or according to facial image In facial image quantity determine.The facial image of selected first preset quantity constitutes first alternative image Group.Since it is desired that the multiple alternative image groups of building, so the first preset quantity is certainly less than the facial image in facial image group Quantity.

It is understood that each facial image corresponds to an acquisition time, and each face figure in facial image group The acquisition time of picture is all different.The facial image of selected first preset quantity, the face that can be connected for acquisition time Image.Because acquisition time is connected, illustrate that corresponding facial image is continuous acquisition, each facial image and a upper people Face image or next facial image have continuity.In general, the difference of the continuous facial image of acquisition time is not too Greatly, the state of the face and in facial image is continuously, to screen mass from the facial image of an alternative image group The probability of best facial image to be utilized is higher.

S204: image group alternative for any non-first determines the facial image group currently non-selected face The destination number of image, it is currently unselected from the facial image group if the destination number is greater than the second preset quantity Facial image in select the facial image of the second preset quantity, and using the facial image of selected second preset quantity and The facial image to be utilized screened from upper one alternative image group constructs an alternative image group；If the number of targets Amount be not more than the second preset quantity, using the facial image group currently non-selected face images and from upper one it is standby The facial image to be utilized that image group is screened is selected, an alternative image group is constructed.

It is understood that destination number can be understood as the number of current non-selected facial image in facial image group Amount, i.e., except the alternately quantity of the facial image in addition to the facial image in image group in current face's image group.The Two preset quantities can be the empirical value of artificial settings, or predetermined this needs to construct in alternative image group The difference of the quantity of facial image and the quantity of the last facial image to be utilized determined.For different alternative image groups, Two preset quantities may be the same or different.The facial image to be utilized screened from each alternative image group can be One, or multiple.It should be noted that the quantity of the facial image to be utilized screened from alternative image group is less than The quantity of facial image in the alternative image group.In addition, the quantity of the facial image in each alternative image group is identical In the case of, the second preset quantity and last the sum of obtained quantity of facial image to be utilized of screening is the first preset quantities.

It in embodiments of the present invention, can be from facial image group currently not when destination number is greater than the second preset quantity The facial image of the second preset quantity is selected in the facial image selected.Specifically, the face of selected second preset quantity The acquisition time of selected facial image before the corresponding acquisition time of image is later than.Acquisition time is selected before being later than The acquisition time of facial image not only can also reduce repetition choosing to avoid the appearance for the case where leakage is selected in facial image group The probability selected.

In addition, the acquisition time of this selected facial image can also be adopted with last selected facial image It is mutually continuous to collect the time.The acquisition time of acquisition time and last selected facial image is mutually continuous, it is ensured that this time Relevance with the time between the alternative image group of building and the alternative image group of last building, two alternative image group tools Having time relevance illustrates that the facial image in the two alternative image groups has relevance.When will be from an alternative image group The facial image to be utilized that middle screening obtains, this alternative figure is constructed with the facial image of the second preset quantity of selection jointly As group can be directed to this structure because facial image to be utilized and the facial image of the second preset quantity have correlation in this way The alternative image group built is arranged unified screening rule, and filters out in the alternative image group constructed from this this best in quality Facial image probability it is higher.

If destination number is not more than the second preset quantity, by non-selected facial image current in facial image group The facial image to be utilized obtained with last screening, constructs an alternative image group.It should be noted that when determining number of targets When amount is 0, the facial image to be utilized that last time screening obtains only has one, then the facial image to be utilized is exactly target face Image selects a facial image, as target if facial image to be utilized is multiple from multiple facial images to be utilized Facial image.

S205: after each alternative image group is constituted, based on the goal gradient of each facial image in the alternative image group, The screening rule for being directed to the alternative image group is determined, and by corresponding screening rule, to the face in the alternative image group Image is screened, and facial image to be utilized is obtained.

S206: based on all facial images to be utilized corresponding to the facial image group, the facial image group is determined Corresponding target facial image.

In the present embodiment, S205-S206 is identical as the S104-S105 of above-described embodiment, and this will not be repeated here.

In embodiments of the present invention, the face figure of last screening obtained facial image to be utilized and this selection is utilized Picture constitutes an alternative image group.Facial image to be utilized is added in next alternative image group, is further sieved Choosing, it is ensured that the facial image to be utilized obtained below is than the high-quality likelihood ratio of the facial image to be utilized obtained before Higher, the selection target facial image from the facial image to be utilized finally obtained can obtain the preferable target person of mass ratio The accuracy rate of recognition of face can be improved compared to the prior art in face image.

Below with reference to another specific embodiment, it is provided for the embodiments of the invention a kind of facial image selection method progress It introduces.

As shown in figure 3, a kind of facial image selection method provided by the embodiment of the present invention, may include steps of:

S301: facial image group is obtained, wherein include multiple facial images in the facial image group.

Wherein, S301 is identical as the S101 of above-described embodiment in the present embodiment, and the associated description content of S301 may refer to The associated description content of S101, this will not be repeated here.

S302: as follows, the goal gradient of each facial image in the facial image group is determined:

Step A1: the first face attribute value of current facial image is determined；The first face attribute value is the first The corresponding value of face attribute, first face character are a face character in the first kind face character.

In embodiments of the present invention, the determination method of the goal gradient of each facial image is the same, for each people Face image can execute the content of step A1 to step A4.

It is understood that current facial image can be understood as in the current facial image for not determining goal gradient One facial image.First face character can be the randomly selected face character from first kind face character, or To the maximum face character of quality of human face image influence degree in first kind face character, certainly, there is also other selections first The method of face character, herein without enumerating.

Step A2: judge whether the first face attribute value meets the corresponding default item of first face character Part, wherein first face character corresponds at least one preset condition, and corresponding to first face character one it is pre- If condition is the attribute-value ranges for the grade setting of the first face character, a preset condition and a grade pair It answers.

It should be noted that a condition is only preset for a grade, for the category of preset condition setting Property value may range from least one attribute value segment, can be a specific attribute value.Illustratively, the first face category Property is clarity, then the corresponding preset condition of one of grade of first face character can be with are as follows: is less than more than or equal to 0 20；First face character is brightness, then the corresponding preset condition of one of grade of first face character can be with are as follows: is greater than It is less than or equal to 255 less than 50 or greater than 200 equal to 0；First face character is positive and negative face, then first face character is corresponding pre- If condition can be with are as follows: negative face.Certainly, above-mentioned examples cited, only a specific example of the embodiment of the present invention do not have to limit and make With.Positive and negative face mentioned here includes positive face and negative face, and positive face refers to that deflection angle is less than the face of first angle threshold value or bows The elevation angle is less than the face of second angle threshold value, it can be understood as than the face of calibration；Negative face refers to that deflection angle is more than or equal to first The face or pitch angle of angle threshold are more than or equal to the face of second angle threshold value, can be big side face, face of bowing greatly, erroneous detection It is non-face.It is understood that deflection angle is one of angle in 3 angles for describe human face posture, it is face in water Square upwards left-right rotation and the angle that is formed.Pitch angle is also one of angle in 3 angles for describe human face posture, Be face rotate upwardly and downwardly in the vertical direction and the angle that is formed.

Step A3: when the judgment result is yes, by the corresponding grade of the preset condition met, it is determined as current face The goal gradient of image.

It is understood that when judging the corresponding preset condition of the first face attribute value the first face character of satisfaction When, by the corresponding grade of the preset condition of satisfaction, it is determined as the goal gradient of current facial image.Determining a face figure After the goal gradient of the face character of picture, other face characters in the facial image first kind face character would not be determined again Corresponding value.

Step A4: when the judgment result is no, a unemployed face is selected from the first kind face character First face character is replaced with selected face character by attribute, is returned and is executed determine current facial image the The step of one face character value.

It is understood that judging result be it is no, illustrate that the first face character pair is not satisfied in the first face attribute value The preset condition answered.When the judgment result is no, it can not determine the goal gradient of the facial image, then it can be from first kind face A unemployed face character is selected in attribute, as the first new face character, and returns to step A1, until really Make the goal gradient of the facial image.

It should be noted that being directed to different facial images, the quantity of the face character value of required determination can be identical, It can be different.

As one embodiment of the present invention, the first kind face character includes positive and negative face, clarity, brightness, screening Gear degree, deflection angle and pitch angle.

It is understood that the corresponding value of each face character in first kind face character, it can be true using classifier It is fixed, it can also be obtained by calculation, herein without limiting.

It should be noted that coverage extent mentioned here refers to the degree blocked to human face region.Coverage extent can be with It is divided into that mouth and nose block, eyes block, seriously block, slightly block, is unobstructed, seriously blocking is except mouth and nose block, eyes block Except shielded area be more than that preset upper limit being blocked, for example, facial image is the image to drink water, because cup has sheltered from people The partial region of face can determine that the facial image is seriously to block, and slightly blocking is in addition to mouth and nose block, eyes block Shielded area is less than blocking for preset upper limit.Certainly, coverage extent is not limited in above-mentioned classification, can pass through shielded area Coverage extent is divided, it can also be by the region division coverage extent etc. that blocks.

S303: the corresponding multiple alternative image groups of the facial image group are constructed, wherein the multiple alternative image group is contained The face images in the facial image group are covered, and each alternative image group includes at least two facial images；

S304: after each alternative image group is constituted, based on the goal gradient of each facial image in the alternative image group, The screening rule for being directed to the alternative image group is determined, and by corresponding screening rule, to the face in the alternative image group Image is screened, and facial image to be utilized is obtained；

S305: based on all facial images to be utilized corresponding to the facial image group, the facial image group is determined Corresponding target facial image.

Wherein, the description content of S303-S305 may refer to the description content with the corresponding portion of above-described embodiment, herein It does not repeat them here.

It in embodiments of the present invention, would not again really after determining the goal gradient of face character of a facial image The corresponding value of other face characters in the fixed facial image first kind face character, can gradually reduce in this way it needs to be determined that people The quantity of the facial image of face attribute value can save resource, and the speed that the goal gradient that can accelerate facial image determines.

As shown in figure 4, a kind of facial image selection method provided by the embodiment of the present invention, may include steps of:

S401: facial image group is obtained, wherein include multiple facial images in the facial image group.

S402: the first kind face character value based on each facial image in the facial image group determines each face The goal gradient of image, wherein the first kind face character value is the corresponding value of first kind face character, the first kind people Face attribute includes at least one face character.

S403: the corresponding multiple alternative image groups of the facial image group are constructed, wherein the multiple alternative image group is contained The face images in the facial image group are covered, and each alternative image group includes at least two facial images.

Wherein, the description content of S401-S403 may refer to the description content with the corresponding portion of above-described embodiment, herein It does not repeat them here.

S404: after each alternative image group is constituted, judge whether the facial image in the alternative image group meets first Default screening conditions, execute S405, otherwise terminate；Wherein, the described first default screening conditions are as follows: the people in the alternative image group The goal gradient of face image is all different, and is not simultaneously the corresponding grade of the second face character；Second face character is A face character in the first kind face character.

As one embodiment of the present invention, each face character in the first kind face character is at least one A grade has correspondence.

It is understood that the goal gradient of facial image can be according to the determination of face character value, it further, can Think face character corresponding grade, therefore, each face character in first kind face character has at least one grade Correspondence.If corresponding grade is not present in a face character in first kind face character, it is determined that this face character Value does not have any contribution to the goal gradient for determining facial image, and determines that this does not have the face character value of corresponding grade, It also wastes time, and increases the time of the goal gradient of determining facial image.

As one embodiment of the present invention, second face character is deflection angle.

It should be noted that experiment based on experience value, in advance or the influence degree to picture quality, can will deflect Angle is as the second face character.

In embodiments of the present invention, a face character is bigger to the influence degree of picture quality, then the face character pair The lower grade answered, conversely, the corresponding higher grade of the face character.Second face character can be in first kind face character To the smallest face character of picture quality influence degree.

It illustratively, include facial image C and facial image D in the alternative image group, the goal gradient of facial image C is The goal gradient of A02, facial image D are A04, and the corresponding grade of the second face character is A07-A13, in the alternative image group The goal gradient of facial image is different, and is not simultaneously any of A07-A13, then may determine that in the alternative image group Facial image meets the first default screening conditions, executes S405.

S405: by pre-set first screening rule, it is determined as the screening rule for the alternative image group, wherein First screening rule is the rule for screening the highest facial image of goal gradient.

It is understood that when the goal gradient of the facial image in alternative image group is not the second face character pair simultaneously The grade answered, and it is not identical when because the second face character is minimum to the influence degree of picture quality, in first kind face character Other face characters it is larger compared to the influence degree of the second face character, then can determine the face in the alternative image group There are biggish differences for image.In conclusion the highest facial image of goal gradient can be screened directly, screening mode is simpler It is single quick, the time of screening can be saved.

In embodiments of the present invention, the first screening rule can be in advance for the default screening conditions setting of satisfaction first Screening rule.

S406: by corresponding screening rule, screening the facial image in the alternative image group, obtains to benefit Use facial image.

Continue the example in S404, screen the highest facial image of goal gradient, the facial image to be utilized of acquisition is behaved Face image D.

S407: based on all facial images to be utilized corresponding to the facial image group, the facial image group is determined Corresponding target facial image.

Wherein, the description content of S406-S407 may refer to the description content with the corresponding portion of above-described embodiment, herein It does not repeat them here.

As one embodiment of the present invention, judging that it is default that the facial image in the alternative image group is unsatisfactory for first In the case where screening conditions, method can also include:

Step B1: the third face attribute value of each facial image in the alternative image group is determined；Wherein, the third party Face attribute value is the corresponding value of third face character, and the third face character is except the face in the first kind face character Face character except attribute.

When the facial image in alternative image group is unsatisfactory for the first default screening conditions, it is thus necessary to determine that the alternative image group In each facial image third face attribute value.Determine that the method for third face attribute value can be by the property of third face character What matter determined.Illustratively, third face character is interpupillary distance, determines that the method for third face attribute value can be with are as follows: according to key Point location algorithm calibrates multiple key points in facial image, positions the pupil of eyes, calculates on Liang Ge pupil center line Pixel quantity, as interpupillary distance.Certainly, above-mentioned examples cited, only a specific example of the embodiment of the present invention do not have limit It is set for using.

As one embodiment of the present invention, the third face character can be interpupillary distance.

In embodiments of the present invention, it is contemplated that the contribution that each face character of facial image screens face, selection Interpupillary distance.

Step B2: judging whether the facial image in the alternative image group meets the second default screening conditions, if it is satisfied, Execute step B3；Wherein, the described second default screening conditions are as follows: the goal gradient of the facial image in the alternative image group is simultaneously For the corresponding grade of second face character and in the difference between the third face attribute value of every two facial image Maximum difference be greater than the quantity of the first preset threshold and third facial image less than the second preset threshold；The third party Face image is the facial image that the third face attribute value is less than third predetermined threshold value.

It should be noted that if the grade of the facial image in an alternative image group is simultaneously the second face character pair The grade answered, it is contemplated that contribution of the third face character to quality of human face image more accurately screens facial image, It then needs further to determine the screening rule for being directed to the alternative image group according to third face attribute value.

Illustratively, the facial image in the alternative image group is facial image A and facial image B, the mesh of facial image A Mark grade is A07, interpupillary distance 50；The goal gradient of facial image B is A08, interpupillary distance 35, it is believed that facial image B is small Interpupillary distance facial image；The corresponding grade of second face character is A07-A13, and the first preset threshold is 10；Second preset threshold is 2, then it can determine that the facial image in the alternative image group meets the second default screening conditions.Certainly, above-mentioned examples cited, only For a specific example of the embodiment of the present invention, do not have restriction effect.

Step B3: by pre-set second screening rule, it is determined as the screening rule for the alternative image group；Its In, second screening rule is the rule for screening the highest facial image of the first comprehensive score.

It is understood that the second screening rule is in advance for the alternative image group setting for meeting the second screening conditions 's.Facial image in each alternative image group is different, and determining goal gradient is different, and goal gradient is different, identified sieve Choosing rule is different, and different screening rules is different to the facial image screening mode in alternative image group.

As one embodiment of the present invention, first comprehensive score is calculated by following two step :

Step 1: obtaining the second class face character of the facial image for each facial image in the alternative image group In each face character reference value, wherein the second class face character includes third face character, the ginseng of face character It is determining based on the corresponding value of face character for examining value.

In embodiments of the present invention, the corresponding value of face character is exactly face character value, in the alternative image group Each facial image, first obtains the second class face character value of the facial image, and the second class face character value is the second class face The corresponding value of attribute, the i.e. corresponding value of each face character in the second class face character.For different face characters, it is somebody's turn to do The mode of the reference value of face character is different, can be based on the face attribute value, by being directed to face character setting in advance Reference value method of determination determines the reference value of the face character.Reference value method of determination for face character setting is to consider The factors such as the property to the face character are pre-set.

As one embodiment of the present invention, the second class face character be interpupillary distance, pitch angle, coverage extent and partially Corner.

In embodiments of the present invention, it is contemplated that influence of the difference of the interpupillary distance of facial image to the screening of facial image, needle To the method for determination of the interpupillary distance of facial image are as follows: based on the difference between facial image interpupillary distance, determine the interpupillary distance of facial image Reference value.It illustratively, include facial image C and facial image D in the alternative image group, the interpupillary distance of facial image C is 32, people The interpupillary distance of face image D is 45, and two facial image interpupillary distance differences are 13, and it is default that interpupillary distance one of two facial images is greater than second Threshold value, one less than the second preset threshold, interpupillary distance influence is bigger, according to the reference value method of determination for being directed to interpupillary distance in advance, really The reference value for determining the interpupillary distance of facial image C is 0, and the reference value of facial image D interpupillary distance is 13.If the interpupillary distance of facial image C is not Become, the interpupillary distance difference of 40, two facial images of interpupillary distance of facial image D is 8, and two facial images are the faces of small interpupillary distance Image, according to for the reference value method of determination of interpupillary distance setting, the reference value of the interpupillary distance of the facial image C of acquisition is 0, people in advance The reference value of face image D interpupillary distance is 16.If the interpupillary distance of facial image C is constant, 60, two face figures of interpupillary distance of facial image D The interpupillary distance difference of picture is 28, and the gap of facial image is bigger, according to the reference value method of determination being arranged in advance for interpupillary distance, is obtained The reference value of the interpupillary distance of the facial image C obtained is 0, and the reference value of facial image D interpupillary distance is 14.Certainly, above-mentioned examples cited, only For a specific example of the embodiment of the present invention, do not have restriction effect.

For defining that deflection angle is bigger, and reference value is smaller in the reference value method of determination of deflection angle setting, specifically really Determine mode are as follows:, will using the reference value of the deflection angle of the deflection angle difference of the two facial images facial image small as deflection angle For the reference value of the deflection angle of another facial image as preset value, which is the deflection angular difference less than two facial images The nonnegative number of value.It is, of course, also possible to by the integral multiple of the deflection angle difference of two facial images, the face figure small as deflection angle The deflection angle difference of two facial images plus a fixed value or can also be subtracted one admittedly by the reference value of the deflection angle of picture Value after definite value, the reference value of the deflection angle of the facial image small as deflection angle.Illustratively, the facial image C of acquisition Deflection angle be 10 degree, the deflection angle of facial image D is 26 degree, and the deflection angle difference of two facial images is 16 degree, acquisition The reference value of the deflection angle of facial image C is 16, and the reference value of the deflection angle of the facial image D of acquisition is 0.

It should be noted that the reference value method of determination for pitch angle can be with reference to the determination side of the reference value of deflection angle Formula, the reference value mode for obtaining the pitch angle of facial image is similar with the mode of reference value of deflection angle of facial image is obtained, This will not be repeated here.

Obtain the mode of the reference value of clarity are as follows: for each facial image in the alternative image group, obtain the face The clarity of image is determined according to the corresponding relationship between pre-set clarity numberical range and the reference value of clarity Clarity numberical range belonging to the clarity of the facial image, by reference corresponding to identified clarity numberical range Value, the reference value of the clarity as the facial image.Illustratively, the value range of clarity is 0-100, then can be preparatory When setting clarity numberical range is 0-20, the reference value of corresponding clarity is 1；When clarity numberical range is 21-40, The reference value of corresponding clarity is 2；When clarity numberical range is 41-60, the reference value of corresponding clarity is 3； When clarity numberical range is 61-80, the reference value of corresponding clarity is 4；When clarity numberical range is 81-100, institute The reference value of corresponding clarity is 5.The clarity of facial image C is 60, then the reference value of the clarity of facial image C is 3, The clarity of facial image D is 80, then the reference value of the clarity of facial image D is 4.

Obtain the mode of the reference value of coverage extent are as follows: for each facial image in the alternative image group, obtain the people The coverage extent of face image, according to the corresponding relationship between pre-set coverage extent and the reference value of coverage extent, by institute Reference value corresponding to determining coverage extent, the reference value of the coverage extent as the facial image.Illustratively, face figure It is in advance 3 for the reference value for seriously blocking setting as the coverage extent of C is seriously to block, then the coverage extent of facial image C Reference value be 3.The coverage extent of facial image D is slightly to block, and is in advance 4 for the reference value for slightly blocking setting, then The reference value of the coverage extent of facial image D is 4.

Step 2: being set for each facial image in the alternative image group according to the third face character is directed in advance The the first weight combination set, is weighted the reference value of each face character of the facial image, obtains the facial image The first comprehensive score.

It should be noted that combine only one for the first weight of third face character setting in advance, and this It include the weight of each face character in the second class face character in the combination of one weight.Weight and the in the combination of first weight Face character in two class face characters is one-to-one.

Illustratively, identified first weight group is combined into 20%, 30%, 15%, 25% and 10%, facial image C's The reference value of face character is respectively 0,16,10,3,3, the reference value of the face character of facial image D is respectively 13,0,0,4, 4, then the first comprehensive score of facial image C is 7.35, and the first comprehensive score of facial image D is 4.Above-mentioned examples cited, only For a specific example of the embodiment of the present invention, do not have restriction effect.

In embodiments of the present invention, the reference value that facial image is determined based on face character value, by different face characters Compare on the same horizontal line, so that calculated first comprehensive score is more accurate, can accurately reflect face figure The quality of picture.

As one embodiment of the present invention, judging that it is default that the facial image in the alternative image group is unsatisfactory for second In the case where screening conditions, method can also include:

Step C1: judging whether the facial image in the alternative image group meets third and preset screening conditions, if it is holding Row step C2；Wherein, the third presets screening conditions are as follows: the goal gradient of the facial image in the alternative image group is described The corresponding grade of second face character, and the grade difference between every two goal gradient is all larger than the 4th preset threshold.

If the facial image in the alternative image group is unsatisfactory for the second default screening conditions, the people in the alternative image Face image is also unsatisfactory for the first default screening conditions, in order to determine the screening rule of the alternative image group, needs further to sentence Whether the facial image to break in the alternative image meets third screening conditions.

Step C2: by first screening rule, it is determined as the screening rule for the alternative image group.

If the goal gradient of the facial image in alternative image group is the corresponding grade of second face character, and every Grade difference between two goal gradients is all larger than the 4th preset threshold, mainly considers the second face character to face figure at this time The influence of the influence of image quality amount, i.e. goal gradient to quality of human face image.If the every two target etc. in the alternative image group Grade difference between grade is all larger than the 4th preset threshold, and the difference in quality of facial image is bigger, then can be directly according to mesh Mark grade is screened, and the highest facial image of goal gradient is screened, screen in this way it is simple and fast, and when can save screening Between.

As one embodiment of the present invention, judging that it is default that the facial image in the alternative image group is unsatisfactory for third In the case where screening conditions, method can also include:

By pre-set third screening rule, it is determined as the screening rule for the alternative image group；Wherein, described Three screening rules are the rule for screening the highest facial image of the second comprehensive score.

In embodiments of the present invention, third screening rule, which can be to be directed in advance, is unsatisfactory for the default screening conditions setting of third Screening rule.Meet the alternative image group that facial image meets third screening rule, can be facial image goal gradient phase Same alternative image group, or facial image is respectively grade etc. adjacent in the corresponding grade of the second face character.

Second comprehensive score is through the following steps that be calculated:

1, the third class face character value of each facial image in the alternative image group is determined；Wherein, the third class people Face attribute value is the corresponding value of third class face character, the third class face character be except the first kind face character and The face character except face character in second class face character.

It should be noted that the face character in third class face character is not belonging to first kind face character, also it is not belonging to Second class face character.Face character in third class face character is to preset.

As one embodiment of the present invention, the third class face character includes whether as yin-yang face, open and close eyes and It opens and shuts up.

It is understood that yin-yang face refers to that the brightness in a part of region of face is greater than a predetermined luminance value, or The brightness in a part of region of person is less than the face of another predetermined luminance value.Illustratively, illumination is mapped on the face of side, is caused The side face is brighter, and other side face is than darker, so that other side face is unclear, there are shades, then can claim this When face be yin-yang face.Certainly, above-mentioned examples cited, only a specific example of the embodiment of the present invention do not have to limit and make With.

For whether yin-yang face, shut up, open and close eyes, can carry out calculating using predetermined computation rule and determine pair The value answered can also determine corresponding value etc. by trained neural network, herein without enumerating.If passed through Trained neural network obtains face character value, if yin-yang face, shut up, these three face characters that open and close eyes respectively correspond The function of a kind of neural network, each neural network is different, but network structure can be the same.

2, the relationship combined according to preset goal gradient with the second weight determines each face figure in the alternative image group Second weight of picture combines.

In embodiments of the present invention, in alternative image group there is a goal gradient in each facial image, according to default The relationship that is combined with the second weight of goal gradient, can determine the second weight combination of each facial image.

3, it for each facial image in the alternative image group, is combined according to the second weight for determining the facial image With each face character value in the 4th class face character value, the second comprehensive score of the facial image is obtained, wherein described Four class face character values are the corresponding value of the 4th class face character, and the 4th class face character includes the first kind face category Face character in property, the second class face character and the third class face character.

As one embodiment of the present invention, the 4th class face character include clarity, brightness, coverage extent, Deflection angle, pitch angle, interpupillary distance, whether be yin-yang face, open and close eyes and open and shut up.

In embodiments of the present invention, the 4th class face character includes the above-mentioned described face character in addition to positive and negative face.

According to each face character value determined in the combination of the second weight and the 4th class face character value, the face is obtained Second comprehensive score of image, the second comprehensive score are obtained based on 8 face attribute values in the 4th class face character value, Each face character value reflects the superiority and inferiority of a face character.Second comprehensive score is overall merit quality of human face image quality Different dimensions index obtain, can reflect the quality of each facial image.

As one embodiment of the present invention, each facial image in the alternative image group, according to institute Determine that each face character value in the combination of the second weight and the 4th class face character value of the facial image is weighted, The step of obtaining the first comprehensive score of the facial image may include:

Step D1: for each facial image in the alternative image group, according to pre-set 4th class face character In each face character it is corresponding value value between mapping relations, determine the 4th class face category of the facial image The value of each face character in property, wherein each face character corresponds to the same value range.

The corresponding value of each face character, since the property of face character is different, the corresponding value of somebody's face attribute compares Greatly, the corresponding value of somebody's face attribute is smaller.In order to more accurately evaluate the quality of each facial image, of the invention real It applies in example, the corresponding value of different faces attribute is normalized to same value range, the corresponding value of different faces attribute is in same model In enclosing, the corresponding value of face characters different in this way can be compared.

Illustratively, value range is 0-9, and interpupillary distance 50 can determine interpupillary distance according to pre-set mapping relations Value is 8.Certainly, above-mentioned examples cited, only a specific example of the embodiment of the present invention do not have restriction effect.

It should be noted that only there are two value, the corresponding value of yin-yang face, non-yin for whether yin-yang face Positive face corresponds to another value, and the corresponding value of yin-yang face is less than the corresponding value of non-yin-yang face；Likewise, being closed for opening For eye, only there are two value, a corresponding value of opening eyes, another corresponding value of closing one's eyes, and corresponding value of opening eyes is big In corresponding value of closing one's eyes；Likewise, for opening and shutting up, only there are two value, a corresponding value of opening one's mouth, correspondence of shutting up Another value, and corresponding value of opening one's mouth is greater than corresponding value of shutting up.

Step D2: it is combined according to the second weight for determining the facial image, each face corresponding to the facial image The value of attribute is weighted, and obtains the second comprehensive score of the facial image.

Here the second comprehensive score of acquisition is identical as the principle of above-mentioned the first comprehensive score of acquisition, and this will not be repeated here.

In embodiments of the present invention, it is contemplated that the concrete condition of the facial image in each alternative image group, pointedly Determine the screening rule in each alternative image group, determining screening rule is more accurate, and then can filter out quality and compare Good facial image to be utilized, improves the accuracy rate of recognition of face.

The facial image selection method provided in the embodiment of the present invention is illustrated below with reference to specific example.

Be illustrated first to the preset condition of the corresponding grade of each face character and each grade: positive and negative face is corresponding Grade is A01, and the preset condition of A01 is negative face；The corresponding grade of clarity is A02, and the preset condition of A02 is that definition values are small In 30；The corresponding grade of brightness is A03, and the preset condition of A03 is definition values less than 55 or more than or equal to 200；Coverage extent Corresponding grade is A04 and A05, and the preset condition of A04 is blocked for mouth and nose, and the preset condition of A05 is blocked for eyes；

The corresponding grade of deflection angle is A06-A13, and the preset condition of A06 is that deflection angle absolute value is greater than 35 or pitch angle It is also the corresponding grade of pitch angle that absolute value, which is greater than 30, A06,；The preset condition of A07 is that deflection angle absolute value is less than or equal to 35 degree Greater than 30 degree；The preset condition of A08 is that deflection angle absolute value is less than or equal to 30 degree greater than 25 degree；The preset condition of A09 is deflection Angle absolute value is less than or equal to 25 degree and is greater than 20 degree；The preset condition of A10 is that deflection angle absolute value is less than or equal to 20 degree greater than 15 degree； The preset condition of A11 is that deflection angle absolute value is less than or equal to 15 degree greater than 10 degree；The preset condition of A12 is that deflection angle absolute value is small It is greater than 5 degree in being equal to 10 degree；The preset condition of A13 is that deflection angle absolute value is less than or equal to 5 degree more than or equal to 0 degree.

Obtaining facial image group A includes 5 facial images, and respectively facial image 1-5, facial image 1-5 are according to adopting The collection time successively sorts.Positive and negative face screening is carried out to facial image 1, determines that facial image is negative face, it is determined that face figure As 1 goal gradient is A01.Positive and negative face screening is carried out to facial image 2, determines that facial image 2 is positive face, it is determined that face figure As 2 definition values, identified definition values are 80；Then determine that the brightness value of facial image 2, identified brightness value are 100；At this time again without the goal gradient of confirmation facial image 2, it is thus necessary to determine that the coverage extent of facial image 2, it is identified Coverage extent is slightly to block, it is also necessary to determine that the deflection angle of facial image 2, deflection angle are 10 degree, it is also necessary to determine face figure The pitch angle of picture 2 is 4 degree, then the goal gradient of identifiable facial image 2 is A12.According to above-mentioned confirmation facial image 2 The method of goal gradient determines that the goal gradient of facial image 3 is A08, and the goal gradient of facial image 4 is A04, facial image 5 goal gradient is A09.

In the goal gradient for having determined 5 facial images, alternative image group 1 is constructed, alternative image group 1 includes facial image 1 and facial image 2, determine that the facial image in alternative image group 1 meets the first default screening conditions, in alternative image 1 Screening rule be the first default screening rule, alternative image group 1 is screened by the first default screening rule, acquisition Image to be utilized is facial image 2.

After to the screening of alternative image 1, alternative image group 2 is constructed using facial image 2 and facial image 3, it is alternative to scheme As the facial image in group 2 is unsatisfactory for the first default screening conditions, it is thus necessary to determine that the interpupillary distance of facial image 2 and facial image 3, people The interpupillary distance of face image 2 is 45, and the interpupillary distance of facial image 3 is 32, and the first preset threshold is 15, and the second preset threshold is 2, and third is pre- If threshold value is 40, alternative image group 2 meets the second screening conditions, the face character reference value of facial image 2 is 13,10,0, 3,4, the face character reference value of facial image 3 is 0,0,5,4,5, and the first weight group is combined into 20%, 30%, 10%, 25% With 15%, the second generic attribute value of the first comprehensive score 6.95 of facial image 2, facial image 3 is respectively 1.95.Obtain to It is facial image 2 using facial image.

After to the screening of alternative image 2, alternative image group 3 is constructed using facial image 2 and facial image 4, it is alternative to scheme As 3 the first default screening conditions of satisfaction of group, then the facial image to be utilized obtained is facial image 2.

After to the screening of alternative image 3, alternative image group 4 is constructed using facial image 2 and facial image 5, it is alternative to scheme The first default screening conditions are unsatisfactory for as organizing the facial image in 4, determine that the interpupillary distance of facial image 5 is 50, then alternative image group 4 In facial image be unsatisfactory for the second default screening conditions.4th preset threshold is 2, then can determine the people in alternative image group Face image meets third and presets screening conditions, it is determined that pre-set first screening rule is the sieve for alternative image group 4 Choosing rule, obtaining facial image to be utilized is facial image 2.At this point, facial image 2 is the target face of facial image group A Image.

Corresponding to above-mentioned embodiment of the method, the embodiment of the invention provides a kind of facial image selection devices, referring to Fig. 5 Shown, described device includes:

Module 501 is obtained, for obtaining facial image group, wherein include multiple facial images in the facial image group；

First determining module 502, for the first kind face character based on each facial image in the facial image group Value, determines the goal gradient of each facial image, wherein the first kind face character value is that first kind face character is corresponding Value, the first kind face character include at least one face character；

Module 503 is constructed, for constructing the corresponding multiple alternative image groups of the facial image group, wherein the multiple Alternative image group covers the face images in the facial image group, and each alternative image group includes at least two faces Image；

First obtains module 504, for being based on each face in the alternative image group after each alternative image group is constituted The goal gradient of image determines the screening rule for being directed to the alternative image group, and by corresponding screening rule, alternative to this Facial image in image group is screened, and facial image to be utilized is obtained；

Second determining module 505, for determining based on all facial images to be utilized corresponding to the facial image group Target facial image corresponding to the facial image group.

As one embodiment of the present invention, the building module 503 may include:

As one embodiment of the present invention, the building module 503, for by the face in the facial image group Image is not divided into multiple alternative image groups；

Alternatively,

Determination unit determines the first face attribute value of current facial image；The first face attribute value is first The corresponding value of face character, first face character are a face character in the first kind face character；

As one embodiment of the present invention, the first kind face character may include positive and negative face, clarity, bright Degree, coverage extent, deflection angle and pitch angle.

As one embodiment of the present invention, each face character in the first kind face character is at least one A grade has correspondence；

Described first obtains module 504, may include:

As one embodiment of the present invention, described device can also include:

Second obtains module, for obtaining the second of the facial image for each facial image in the alternative image group The reference value of each face character in class face character, wherein the second class face character includes third face character, and one The reference value of a face character is determined based on the face attribute value；

As one embodiment of the present invention, described device can also include:

As one embodiment of the present invention, the described 4th obtains module, may include:

As one embodiment of the present invention, second face character can be deflection angle；

The third face character can be interpupillary distance；

The second class face character may include interpupillary distance, pitch angle, coverage extent and deflection angle；

The third class face character may include whether as yin-yang face, open and close eyes and open and shut up；

The 4th class face character may include clarity, brightness, coverage extent, deflection angle, pitch angle, interpupillary distance, be It is no for yin-yang face, open and close eyes and open and shut up.

The embodiment of the invention also provides a kind of computer equipments, as shown in fig. 6, including processor 601, communication interface 602, memory 603 and communication bus 604, wherein processor 601, communication interface 602, memory 603 pass through communication bus 604 complete mutual communication,

Memory 603, for storing computer program；

Processor 601 when for executing the program stored on memory 603, realizes following steps:

The specific implementation of facial image selection method performed by above-mentioned computer equipment and preceding method embodiment In the various implementations that refer to it is identical, which is not described herein again.

The communication bus that above-mentioned computer equipment is mentioned can be Peripheral Component Interconnect standard (Peripheral Component Interconnect, PCI) bus or expanding the industrial standard structure (Extended Industry Standard Architecture, EISA) bus etc..The communication bus can be divided into address bus, data/address bus, control bus etc..For just It is only indicated with a thick line in expression, figure, it is not intended that an only bus or a type of bus.

Communication interface is for the communication between above-mentioned computer equipment and other equipment.

Memory may include random access memory (Random Access Memory, RAM), also may include non-easy The property lost memory (Non-Volatile Memory, NVM), for example, at least a magnetic disk storage.Optionally, memory may be used also To be storage device that at least one is located remotely from aforementioned processor.

Above-mentioned processor can be general processor, including central processing unit (Central Processing Unit, CPU), network processing unit (Network Processor, NP) etc.；It can also be digital signal processor (Digital Signal Processing, DSP), it is specific integrated circuit (Application Specific Integrated Circuit, ASIC), existing It is field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete Door or transistor logic, discrete hardware components.

In another embodiment of the present invention, a kind of computer readable storage medium, computer-readable storage are additionally provided Dielectric memory contains computer program, and any face in above-described embodiment is realized when computer program is executed by processor Image-selecting method.

It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that There is also other identical elements in process, method, article or equipment including the element.

Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device reality For applying example, since it is substantially similar to the method embodiment, so being described relatively simple, related place is referring to embodiment of the method Part explanation.

The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the scope of the present invention.It is all Any modification, equivalent replacement, improvement and so within the spirit and principles in the present invention, are all contained in protection scope of the present invention It is interior.

Claims

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

Based on the first kind face character value of each facial image in the facial image group, the target of each facial image is determined Grade, wherein the first kind face character value is the corresponding value of first kind face character, and the first kind face character includes At least one face character；

Construct the corresponding multiple alternative image groups of the facial image group, wherein the multiple alternative image group covers the people Face images in face image group, and each alternative image group includes at least two facial images；

After each alternative image group is constituted, based on the goal gradient of each facial image in the alternative image group, determination is directed to The screening rule of the alternative image group, and by corresponding screening rule, the facial image in the alternative image group is carried out Screening, obtains facial image to be utilized；

Based on all facial images to be utilized corresponding to the facial image group, mesh corresponding to the facial image group is determined Mark facial image.

2. the method according to claim 1, wherein the building facial image group is corresponding multiple alternative The step of image group, comprising:

Image group alternative for any non-first determines the target of the current non-selected facial image of the facial image group Quantity, if the destination number is greater than the second preset quantity, from the current non-selected facial image of the facial image group The facial image of the second preset quantity of middle selection, and utilize the facial image of selected second preset quantity and standby from upper one The facial image to be utilized that image group is screened is selected, an alternative image group is constructed；If the destination number is no more than the Two preset quantities currently non-selected face images and are sieved from upper one alternative image group using the facial image group Obtained facial image to be utilized is selected, an alternative image group is constructed.

3. the method according to claim 1, wherein the building facial image group is corresponding multiple alternative The step of image group, comprising:

Alternatively,

4. the method according to claim 1, wherein described based on each facial image in the facial image group First kind face character value, the step of determining the goal gradient of each facial image, comprising:

Determine the first face attribute value of current facial image；The first face attribute value is that the first face character is corresponding Value, first face character are a face character in the first kind face character；

Judge whether the first face attribute value meets the corresponding preset condition of first face character, wherein institute It states the first face character and corresponds at least one preset condition, and a preset condition corresponding to first face character is needle To the attribute-value ranges of the grade setting of the first face character, a preset condition is corresponding with a grade；

When the judgment result is yes, by the corresponding grade of the preset condition met, it is determined as the target of current facial image Grade；

When the judgment result is no, a unemployed face character is selected from the first kind face character, it will be described First face character replaces with selected face character, returns to the first face attribute value for executing and determining current facial image The step of.

5. according to the method described in claim 4, it is characterized in that, the first kind face character include positive and negative face, clarity, Brightness, coverage extent, deflection angle and pitch angle.

6. method according to claim 1-5, which is characterized in that each people in the first kind face character Face attribute has correspondence at least one grade；

The goal gradient based on each facial image in the alternative image group determines that the screening for the alternative image group is advised Then the step of, comprising:

Judge whether the facial image in the alternative image group meets the first default screening conditions, wherein the described first default sieve Select condition are as follows: the goal gradient of the facial image in the alternative image group is all different, and is not simultaneously the second face character pair The grade answered；Second face character is a face character in the first kind face character；

If it is satisfied, pre-set first screening rule is determined as the screening rule for the alternative image group, wherein First screening rule is the rule for screening the highest facial image of goal gradient.

7. according to the method described in claim 6, it is characterized in that, judging that the facial image in the alternative image group is unsatisfactory for In the case where first default screening conditions, the method also includes:

Determine the third face attribute value of each facial image in the alternative image group；Wherein, the third face attribute value is The corresponding value of third face character, the third face character are in addition to the face character in the first kind face character Face character；

Judge whether the facial image in the alternative image group meets the second default screening conditions, wherein the described second default sieve Select condition are as follows: the goal gradient of the facial image in the alternative image group be simultaneously the corresponding grade of second face character, And for every two facial image third face attribute value between difference in maximum difference be greater than the first preset threshold, with And the quantity of third facial image is less than the second preset threshold；The third facial image is that the third face attribute value is less than The facial image of third predetermined threshold value；

If it is satisfied, pre-set second screening rule is determined as the screening rule for the alternative image group；Wherein, Second screening rule is the rule for screening the highest facial image of the first comprehensive score；

First comprehensive score is through the following steps that be calculated:

For each facial image in the alternative image group, each face in the second class face character of the facial image is obtained The reference value of attribute, wherein the second class face character includes third face character, and the reference value of face character is to be based on being somebody's turn to do The corresponding value determination of face character；

For each facial image in the alternative image group, according to the first weight for being directed to third face character setting in advance Combination, is weighted the reference value of each face character of the facial image, and the first synthesis for obtaining the facial image is commented Point.

8. the method according to the description of claim 7 is characterized in that judging that the facial image in the alternative image group is unsatisfactory for In the case where second default screening conditions, the method also includes:

Judge whether the facial image in the alternative image group meets third and preset screening conditions, wherein the default sieve of the third Select condition are as follows: the goal gradient of the facial image in the alternative image group is the corresponding grade of second face character, and every Grade difference between two goal gradients is all larger than the 4th preset threshold；

9. according to the method described in claim 8, it is characterized in that, judging that the facial image in the alternative image group is unsatisfactory for In the case that third presets screening conditions, the method also includes:

By pre-set third screening rule, it is determined as the screening rule for the alternative image group；Wherein, the third sieve Choosing rule is the rule of the screening highest facial image of the second comprehensive score；

Second comprehensive score is through the following steps that be calculated:

Determine the third class face character value of each facial image in the alternative image group；Wherein, the third class face character Value is the corresponding value of third class face character, and the third class face character is except the first kind face character and the second class The face character except face character in face character；

According to the relationship that preset goal gradient is combined with the second weight, of each facial image in the alternative image group is determined The combination of two weights；

For each facial image in the alternative image group, combined and the 4th according to the second weight for determining the facial image Each face character value in class face character value, obtains the second comprehensive score of the facial image, wherein the 4th class people Face attribute value is the corresponding value of the 4th class face character, and the 4th class face character includes the first kind face character, institute State the face character in the second class face character and the third class face character.

10. according to the method described in claim 9, it is characterized in that, each face figure in the alternative image group Picture is combined according to the second weight for determining the facial image and is carried out with each face character value in the 4th class face character value Weighted calculation, the step of obtaining the second comprehensive score of the facial image, comprising:

For each facial image in the alternative image group, according to each face in pre-set 4th class face character Mapping relations between the corresponding value of attribute and value, determine each face in the 4th class face character of the facial image The value of attribute, wherein each face character corresponds to the same value range；

Combined according to the second weight for determining the facial image, the value of each face character corresponding to the facial image into Row weighted calculation obtains the second comprehensive score of the facial image.

11. according to the method described in claim 10, it is characterized in that, second face character is deflection angle；

The third face character is interpupillary distance；

Whether the 4th class face character includes clarity, brightness, coverage extent, deflection angle, pitch angle, interpupillary distance, is yin-yang Face opens and closes eyes and opens and shuts up.

12. a kind of facial image selection device, which is characterized in that described device includes:

First determining module is determined for the first kind face character value based on each facial image in the facial image group The goal gradient of each facial image, wherein the first kind face character value is the corresponding value of first kind face character, described First kind face character includes at least one face character；

Module is constructed, for constructing the corresponding multiple alternative image groups of the facial image group, wherein the multiple alternative image Group covers the face images in the facial image group, and each alternative image group includes at least two facial images；

First obtains module, is used for after each alternative image group is constituted, based on each facial image in the alternative image group Goal gradient determines the screening rule for being directed to the alternative image group, and by corresponding screening rule, to the alternative image group In facial image screened, obtain facial image to be utilized；

Second determining module, for determining the people based on all facial images to be utilized corresponding to the facial image group Target facial image corresponding to face image group.

13. device according to claim 12, which is characterized in that the building module, comprising:

First building submodule, for selecting the facial image of the first preset quantity, building first from the facial image group A alternative image group；

Second building submodule, be used for image group alternative for any non-first, determine the facial image group currently not by The destination number of the facial image of selection is worked as if the destination number is greater than the second preset quantity from the facial image group The facial image of the second preset quantity is selected in preceding non-selected facial image, and utilizes selected second preset quantity Facial image and the facial image to be utilized screened from upper one alternative image group construct an alternative image group；If The destination number be not more than the second preset quantity, using the facial image group currently non-selected face images and The facial image to be utilized screened from upper one alternative image group constructs an alternative image group.

14. device according to claim 12, which is characterized in that the building module is specifically used for the face figure As the facial image in group is not divided into multiple alternative image groups；

Alternatively,

15. device according to claim 12, which is characterized in that described first obtains module, is used for as follows, Determine the goal gradient of each facial image in the facial image group:

16. device according to claim 15, which is characterized in that the first kind face character includes positive and negative face, clear Degree, brightness, coverage extent, deflection angle and pitch angle.

17. the described in any item devices of 2-16 according to claim 1, which is characterized in that every in the first kind face character One face character has correspondence at least one grade；

Described first obtains module, comprising:

Judging submodule, for judging whether the facial image in the alternative image group meets the first default screening conditions, wherein The first default screening conditions are as follows: the goal gradient of the facial image in the alternative image group is all different, and is not simultaneously The corresponding grade of second face character；Second face character is a face character in the first kind face character；

First determine submodule, for the judging result of the judging submodule be meet in the case where, will be pre-set First screening rule is determined as the screening rule for the alternative image group, wherein first screening rule is screening target The rule of the highest facial image of grade.

18. device according to claim 17, which is characterized in that described device further include:

Third determining module is to determine the alternative figure in ungratified situation for the judging result in the judging submodule As the third face attribute value of facial image each in group；Wherein, the third face attribute value is corresponding for third face character Value, the third face character is face character in addition to the face character in the first kind face character；

First judgment module, for judging whether the facial image in the alternative image group meets the second default screening conditions, In, the second default screening conditions are as follows: the goal gradient of the facial image in the alternative image group is simultaneously second people Maximum difference in the corresponding grade of face attribute and the difference being directed between the third face attribute value of every two facial image is big In the first preset threshold and the quantity of third facial image less than the second preset threshold；The third facial image is described Third face attribute value is less than the facial image of third predetermined threshold value；

4th determining module, for will preset in the case where the judging result of first judging submodule is to meet The second screening rule, be determined as the screening rule for the alternative image group；Wherein, second screening rule is screening the The rule of the highest facial image of one comprehensive score；

Second obtains module, for obtaining the second class people of the facial image for each facial image in the alternative image group The reference value of each face character in face attribute, wherein the second class face character includes third face character, face category Property reference value be determining based on the corresponding value of the face character；

Third obtains module, for being directed to each facial image in the alternative image group, according in advance for the third face The first weight combination of attribute setting, is weighted the reference value of each face character of the facial image, obtains the people First comprehensive score of face image.

19. device according to claim 18, which is characterized in that described device further include:

Second judgment module is to judge that this is alternative in ungratified situation for the judging result in the first judgment module Whether the facial image in image group, which meets third, is preset screening conditions, wherein the third presets screening conditions are as follows: this is alternative The goal gradient of facial image in image group is the corresponding grade of second face character, and between every two goal gradient Grade difference be all larger than the 4th preset threshold；

5th determining module, in the case where the judging result of such as figure judgment module is to be, described first to be sieved Choosing rule, is determined as the screening rule for the alternative image group.

20. device according to claim 19, which is characterized in that described device further include:

6th determining module is that will preset in ungratified situation for the judging result in second judgment module Third screening rule, be determined as the screening rule for the alternative image group；Wherein, the third screening rule is screening the The rule of the highest facial image of two comprehensive scores；

7th determining module, for determining the third class face character value of each facial image in the alternative image group；Wherein, institute Stating third class face character value is the corresponding value of third class face character, and the third class face character is except the first kind people The face character except face character in face attribute and the second class face character；

8th determining module, the relationship for being combined according to preset goal gradient with the second weight determine the alternative image group In each facial image the second weight combination；

4th obtains module, for for each facial image in the alternative image group, according to determining the facial image Each face character value in the combination of second weight and the 4th class face character value, the second synthesis for obtaining the facial image are commented Point, wherein the 4th class face character value is the corresponding value of the 4th class face character, and the 4th class face character includes institute State the face character in first kind face character, the second class face character and the third class face character.

21. device according to claim 20, which is characterized in that the described 4th obtains module, comprising:

Second determines submodule, each facial image for being directed in the alternative image group, according to pre-set 4th class The corresponding value of each face character in face character and the mapping relations between value, determine the described 4th of the facial image The value of each face character in class face character, wherein each face character corresponds to the same value range；

Submodule is obtained, is combined according to the second weight for determining the facial image, each face corresponding to the facial image The value of attribute is weighted, and obtains the second comprehensive score of the facial image.

22. side's device according to claim 21, which is characterized in that second face character is deflection angle；

The third face character is interpupillary distance；

23. a kind of computer equipment, which is characterized in that including processor, communication interface, memory and communication bus, wherein institute Processor, the communication interface are stated, the memory completes mutual communication by the communication bus；

The memory, for storing computer program；

The processor when for executing the program stored on the memory, realizes that claim 1-11 is any described Method and step.