WO2016058520A1 - 人脸图片人名识别方法和装置 - Google Patents

人脸图片人名识别方法和装置 Download PDF

Info

Publication number
WO2016058520A1
WO2016058520A1 PCT/CN2015/091869 CN2015091869W WO2016058520A1 WO 2016058520 A1 WO2016058520 A1 WO 2016058520A1 CN 2015091869 W CN2015091869 W CN 2015091869W WO 2016058520 A1 WO2016058520 A1 WO 2016058520A1
Authority
WO
WIPO (PCT)
Prior art keywords
name
text
person
score
face
Prior art date
Application number
PCT/CN2015/091869
Other languages
English (en)
French (fr)
Inventor
薛红霞
陶哲
胡金辉
Original Assignee
北京奇虎科技有限公司
奇智软件(北京)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京奇虎科技有限公司, 奇智软件(北京)有限公司 filed Critical 北京奇虎科技有限公司
Publication of WO2016058520A1 publication Critical patent/WO2016058520A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172Classification, e.g. identification
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G06F16/5846Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using extracted text
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition

Definitions

  • the present invention relates to the field of computer technology, and in particular to a method and apparatus for recognizing a face picture.
  • face recognition technology is gradually developing, but because face images often have problems in terms of light, angle, etc., it is easy to cause a person name that is difficult to accurately recognize the person corresponding to the face.
  • the present invention has been made in order to provide a face picture person name recognition method and apparatus that overcomes the above problems or at least partially solves the above problems.
  • a method for recognizing a face picture name includes: extracting a person name from text corresponding to a target face picture; performing face recognition on the target face picture to identify the target a similar face picture of the face picture, and obtaining the name of the face in the similar face picture; determining the target face according to the name of the person in the text and the name of the face in the similar face picture The name of the person in the picture.
  • a face picture person name recognition apparatus comprising: a person name extraction module, configured to extract a person name from text corresponding to a target face picture; a face recognition module, configured to Performing face recognition on the target face image, identifying a similar face image of the target face image, and acquiring a name of a face in the similar face image; a name determination module for using the name of the person in the text And a name of a person face in the similar face picture, and determining a person name of the face in the target face picture.
  • a computer program comprising computer readable code that, when executed on a computing device, causes the computing device to perform the face picture name recognition method described above.
  • a computer readable medium wherein the computer program described above is stored.
  • the face picture name recognition method and apparatus of the present invention have at least the following advantages:
  • the face recognition technology is used to identify the target face image, but also the name of the person included in the text corresponding to the target face image is considered, because the text corresponding to the target face image is It often has a close relationship with the target face image, so the name of the person involved in the text contains the name of the face in the target face image, and the name of the similar face recognized by the face recognition technology and the corresponding name
  • the comprehensive consideration of the names obtained in the text will undoubtedly more accurately determine the names of the faces in the target face images.
  • FIG. 1 shows a flow chart of a face picture person name recognition method according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram showing the principle of a face picture person name recognition method according to an embodiment of the present invention
  • FIG. 3 illustrates a partial flowchart of a face picture person name recognition method according to an embodiment of the present invention
  • FIG. 4 is a partial flow chart showing a face picture person name recognition method according to an embodiment of the present invention.
  • FIG. 5 is a block diagram showing a face picture person name recognition apparatus according to an embodiment of the present invention.
  • FIG. 6 is a block diagram showing a module of a face picture person name recognition apparatus according to an embodiment of the present invention.
  • FIG. 7 is a block diagram showing a module of a face picture person name recognition apparatus according to an embodiment of the present invention.
  • Figure 8 is a block diagram schematically showing a computing device for performing a face picture person name recognition method according to the present invention.
  • Fig. 9 schematically shows a storage unit for holding or carrying program code implementing the face picture person name recognition method according to the present invention.
  • an embodiment of the present invention provides a method for recognizing a human face picture name, including:
  • Step 110 Extract a person name from the text corresponding to the target face image.
  • the text may be the news content of the news message; when the target face image is located in a technical document In the middle, the text can be the context content of the target face image.
  • the target face image usually has a close relationship with the corresponding text, so it can be determined that the name of the face in the target face image is likely to exist in the text, for example, a news message is provided.
  • the news content reports the star's concert situation, the news content provides the name of the star in the star photo;
  • a technical document provides a photo of the scientist, and the text below the photo is for the The introduction of the scientist's biography, the name of the scientist is provided in the text below the photo.
  • the number of people extracted in the text is one or more.
  • Step 120 Perform face recognition on the target face image, identify a similar face image of the target face image, and obtain a person name of the face in the similar face image.
  • the similar face image of the target face image is recognized based on the face recognition technology, and the face recognition technology adopted is not limited, and any existing face recognition technology can be used in the implementation.
  • the technical solution of the example a large number of collected face images and a person name of a face in each face image may be pre-stored in the database, and a similar face image may be found from the database through the face recognition technology, and can be determined.
  • the number of names obtained by the face recognition technology is one or more.
  • Step 130 Determine the name of the face of the face in the target face picture according to the name of the person in the text and the name of the face in the similar face picture.
  • the name of the person recognized by the face recognition technology and the name of the person included in the text are comprehensively considered, the name of the face of the face in the target face picture can be more accurately determined.
  • there is no limitation on how to determine the name of the face in the target face image For example, for a face image in a news message, "Zhang San" is extracted from the news content of the news message.
  • Li Si two names, through the face recognition method to obtain two similar face images of the target face image, and determine that the names of the two similar face images are "Li Si, Wang Wu", then The accuracy of the "Li Si” obtained by the two methods is relatively high, so the “Li Si” is determined as the name of the face in the target face image; and, if A person name of "Zhang San” was extracted from the news content of the news message, and a similar face image of the target face image was obtained by the face recognition method, and the name of the similar face image was determined as "Li Si” The result obtained by the face recognition technology is taken as the standard, and the "Li Si" is determined as the face image in the target face picture.
  • the name of the person involved in the text includes the name of the face in the target face image, and Considering the name of a similar face recognized by the face recognition technology and the name of the person obtained from the corresponding text, as shown in the schematic diagram of FIG. 2, it is undoubtedly more accurate to determine the name of the face in the target face picture.
  • another embodiment of the present invention provides a method for recognizing a human face picture name, and the step 130 specifically includes:
  • Step 131 Calculate a first score for the name of the person in the text according to the relationship between the attribute of the text and/or the text and the name of the person in the text.
  • the different attributes of the text, the different relationship between the text and the name of the person in the text can reflect the possibility of the person's name as the person's name in the target face picture. For example, when the attribute of the text is the type of the text, for example, a news message provides a personal photo of the star, and the news content reports the performance of the star, the name of the news content is likely to be a star.
  • the name of each person in the text the first score of each person's name can be set to 90 points; a literary work provides a photo of the author, and the name of the person contained in the text may be only the protagonist of the work, not the name of the author. Then the first score of each person's name in the text can be set to 10 points.
  • the relationship between the text and the name of the text is the number of times the name appears in the text, for example, in a news message with a face image, the name 1 appears 9 times, and the name 2 appears once, it is easy It is understood that the person name 1 is more likely to be the person name of the face in the face picture, so the first point value of the person name 1 can be set to 80 points, and the first point value of the person name 2 can be set to 20 points. In this embodiment, there is no restriction on the relationship between the attributes of the text, the text, and the names of the characters in the text.
  • Step 132 Calculate a second score for the name of the face in the similar face image according to the similarity between the similar face image and the target face image.
  • the target face image with higher similarity should be Has a higher score.
  • the second score is calculated:
  • Step 133 Calculate the result score according to the first score and the second score that the same person name has, and determine the name of the face in the target face image according to the size of the result score.
  • both the first score and the second score reflect the possibility that the obtained person name is the name of the face in the target face image, so the first score and the second score are combined.
  • the score of the result is more likely to reflect whether each person's name is the name of the person's face in the target face image. Then, according to the size of the result score, the corresponding person's name is selected as the name of the face in the target face image, and the accuracy is accurate. very high.
  • the calculation method of the result score is not limited, for example, it may be the addition or multiplication of the first score and the second score.
  • the first score of person name 1 is 80 points
  • the second score is 60 points
  • the first score of person name 2 is 90 points
  • the second score is 20 points
  • the first score of person name 3 is only 50 points.
  • the result score is calculated by adding the first score and the second score.
  • the result score of the person name 1 is 140 points
  • the result score of the person name 2 is 110 points
  • the result score of the person name 3 is 50 points.
  • the first score obtained based on the attribute of the text, the relationship between the text and the person name in the text, and the second score obtained based on the similarity between the similar face image and the target face image Both reflect the possibility of each person's name as the face of the face in the target face image, so the result score is obtained by combining the first score and the second score to select the person's name as the face in the target face image.
  • the name of the person is very accurate. It is especially suitable for the case where multiple candidate names are obtained through text extraction and face recognition technology.
  • step 131 provides a method for recognizing a human face picture name, and step 131 includes:
  • the first sub-score is calculated according to the attribute of the text for the person name in the text
  • the second sub-score is calculated according to the relationship between the text and the person name in the text
  • the first sub-value is calculated according to the first sub-score and the second sub-score.
  • the relationship between the attribute of the text, the text and the name of the person in the text is considered at the same time, so the calculated first score can obviously reflect the possibility of each person being named as the name of the face in the target face picture.
  • the position that appears will reflect whether the name of the person is the name of the person's face in the target face image.
  • the first person name in the news message is often the name of the face in the face image, and later It is less likely that the person whose position appears to be the name of the face in the target face picture is low, so the first sub-score can also reflect whether the name of the person is the name of the person's face in the target face picture.
  • a face image may have multiple texts, and the type may be the title, content, and surrounding text of the face image; the type of the text is different, which can reflect the target person.
  • the ability of the face name of the face in the face image is also different.
  • the title corresponding to a face image is “Star A, Star B is blocked by an international star”, and only the name of the person whose face image may be included can be obtained from the title;
  • "China's new generation popular popular star A" can find the name of the face corresponding to this picture. Therefore, the second sub-score can also reflect whether the name of the person is the name of the person's face in the target face picture.
  • the first name of the person name Name i is calculated as follows:
  • the first score of the person can be standardized:
  • weight p is the first sub-score and weight position is the second sub-score.
  • Another embodiment of the present invention provides a face picture person name recognition method.
  • the attributes of the text include, but are not limited to, the type of the text, the location of the text, and/or the publisher of the text.
  • the type of the text for example, the person in the title is more likely to be the person name of the face in the target face picture, and the person in the body name is the person name of the face in the target face picture.
  • the text is located on a well-known website, the person name is the name of the face in the target face image is higher, and the text is located
  • Non-famous websites are less likely to be named as faces in the face image of the target face; for publishers of text, the possibility of posting the name of the person in the target face image to the person in the text is more authoritative People with higher sexuality, not those published by the authority, are less likely to be named as faces in the face image.
  • the relationship between the text and the name of the person in the text includes, but is not limited to, the position and/or number of occurrences of the person's name in the text in the text.
  • the position of the person's name in the text in the text for example, the person in the upper position in the text is more likely to be the person name of the face in the target face picture, and the position in the text is later. It is less likely that a person will be named as a person's face in the target face image.
  • another embodiment of the present invention provides a method for recognizing a human face picture name, and the step 120 specifically includes:
  • Step 121 Acquire features of the collected face image, and acquire features of the target face image.
  • the position of the face in the face image is detected, and the related features of the face are extracted, and the plurality of features form a multi-dimensional vector for comparison, for example, , 400-dimensional vector.
  • Step 122 Compare the feature of the collected face picture with the feature of the target face picture, and determine a similar face picture according to the comparison result.
  • a manner of effectively recognizing similar face pictures is provided by means of feature comparison.
  • Another embodiment of the present invention provides a method for recognizing a human face picture name, and the step 130 specifically includes:
  • the first weight and the second weight reflect the importance degree of the text and the face recognition. If the first weight is set larger, the user values the result of the text extraction, such as If the second weight is set larger, the user is more concerned with the result of face recognition; therefore, if the credibility of the text source is higher, the first weight may be set larger, if the face recognition algorithm Preferably, the second weight can be set to be larger.
  • the technical solution in this embodiment is advantageous for the user to adjust the influence degree of the first score and the second score, thereby obtaining the most reasonable result score.
  • the result score is calculated for the person name Name i :
  • the person name is output as the person name of the face in the target face picture.
  • Another embodiment of the present invention provides a face picture person name recognition method, and the text includes, but is not limited to, a title, a body, and/or a surrounding text of a target face picture in a document corresponding to the target face picture.
  • FIG. 5 another embodiment of the present invention provides a face picture person name recognition apparatus, including:
  • the person name extraction module 510 is configured to extract a person name from the text corresponding to the target face picture.
  • the text may be the news content of the news message; when the target face image is located in a technical document In the middle, the text can be the context content of the target face image.
  • the target face image usually has a close relationship with the corresponding text, so it can be determined that the name of the face in the target face image is likely to exist in the text, for example, a news message is provided.
  • the news content reports the star's concert situation, the news content provides the name of the star in the star photo;
  • a technical document provides a photo of the scientist, and the text below the photo is for the The introduction of the scientist's biography, the name of the scientist is provided in the text below the photo.
  • the number of people extracted in the text is one or more.
  • the face recognition module 520 is configured to perform face recognition on the target face image, identify a similar face image of the target face image, and obtain a person name of the face in the similar face image.
  • the similar face image of the target face image is recognized based on the face recognition technology, and the face recognition technology adopted is not limited, and any existing face recognition technology can be used in the implementation.
  • the technical solution of the example In this embodiment, a large number of collected face images and a person name of a face in each face image may be pre-stored in the database, and a similar face image may be found from the database through the face recognition technology, and can be determined.
  • the number of names obtained by the face recognition technology is one or more.
  • the name determination module 530 is configured to: according to the name of the person in the text, and the name of the person in the face of the similar face image, Determine the name of the person's face in the target face image.
  • the name of the person recognized by the face recognition technology and the name of the person included in the text are comprehensively considered, the name of the face of the face in the target face picture can be more accurately determined.
  • there is no limitation on how to determine the name of the face in the target face image For example, for a face image in a news message, "Zhang San" is extracted from the news content of the news message.
  • Li Si two names, through the face recognition method to obtain two similar face images of the target face image, and determine that the names of the two similar face images are "Li Si, Wang Wu", then The accuracy of the "Li Si” obtained by the two methods is relatively high, so the “Li Si” is determined as the name of the face in the target face picture; and if, the "Yan” is extracted from the news content of the news message
  • the third person name is a similar face image of the target face image obtained by the face recognition method, and the name of the similar face image is determined as “Li Si”, and the result obtained by the face recognition technology is subject to , "Li Si" is determined as the face image in the target face picture.
  • the name of the person involved in the text includes the name of the face in the target face image, and Considering the name of a similar face recognized by the face recognition technology and the name of the person obtained from the corresponding text, as shown in the schematic diagram of FIG. 2, it is undoubtedly more accurate to determine the name of the face in the target face picture.
  • another embodiment of the present invention provides a face picture name recognition apparatus, and the name determination module 530 includes:
  • the first score calculation module 531 is configured to calculate a first score for the name of the person in the text according to the attribute of the text and/or the relationship between the text and the name of the person in the text.
  • the different attributes of the text, the different relationship between the text and the name of the person in the text can reflect the possibility of the person's name as the person's name in the target face picture. For example, when the attribute of the text is the type of the text, for example, a news message provides a personal photo of the star, and the news content reports the performance of the star, the name of the news content is likely to be a star.
  • the name of each person in the text the first score of each person's name can be set to 90 points; a literary work provides a photo of the author, and the name of the person contained in the text may be only the protagonist of the work, not the name of the author. Then the first score of each person's name in the text can be set to 10 points.
  • the relationship between the text and the name of the text is the number of times the name appears in the text, for example, in a news message with a face image, the name 1 appears 9 times, and the name 2 appears once, it is easy It is understood that the person name 1 is more likely to be the person name of the face in the face picture, so the first point value of the person name 1 can be set to 80 points, and the first point value of the person name 2 can be set to 20 points. In this embodiment, there is no restriction on the relationship between the attributes of the text, the text, and the names of the characters in the text.
  • the second score calculation module 532 is configured to calculate a second score for the person name of the face in the similar face image according to the similarity between the similar face image and the target face image.
  • the target face image with higher similarity should be Has a higher score.
  • the second score is calculated:
  • the result score calculation module 533 is configured to calculate a result score according to the first score and the second score that the same person name has, and determine the name of the face in the target face image according to the size of the result score.
  • both the first score and the second score reflect the possibility that the obtained person name is the name of the face in the target face image, so the first score and the second score are combined.
  • the score of the result is more likely to reflect whether each person's name is the name of the person's face in the target face image. Then, according to the size of the result score, the corresponding person's name is selected as the name of the face in the target face image, and the accuracy is accurate. very high.
  • the calculation method of the result score is not limited, for example, it may be the addition or multiplication of the first score and the second score.
  • the first score of person name 1 is 80 points
  • the second score is 60 points
  • the first score of person name 2 is 90 points
  • the second score is 20 points
  • the first score of person name 3 is only 50 points.
  • the result score is calculated by adding the first score and the second score.
  • the result score of the person name 1 is 140 points
  • the result score of the person name 2 is 110 points
  • the result score of the person name 3 is 50 points.
  • the first score obtained based on the attribute of the text, the relationship between the text and the person name in the text, and the second score obtained based on the similarity between the similar face image and the target face image Both reflect the possibility of each person's name as the face of the face in the target face image, so the result score is obtained by combining the first score and the second score to select the person's name as the face in the target face image. Name of person, this way The accuracy is very high, especially for the case of multiple candidate names obtained through text extraction and face recognition technology.
  • the first point value calculation module 531 calculates a first child score for a person name in the text according to an attribute of the text, and calculates a relationship between the text and the person name in the text.
  • the second sub-score calculates a first score based on the first sub-score and the second sub-score.
  • the relationship between the attribute of the text, the text and the name of the person in the text is considered at the same time, so the calculated first score can obviously reflect the possibility of each person being named as the name of the face in the target face picture.
  • the position where the person's name appears reflects the possibility that the person's name is the name of the person's face in the target face image, such as in a news message.
  • the first person name that appears first is the name of the face in the face image, and the person who appears in the later position is less likely to be the name of the face in the target face image, so the first child score is also It is possible to reflect whether the name of the person is the name of the person's face in the target face picture.
  • a face image may have multiple texts, and the type may be the title, content, and surrounding text of the face image; the type of the text is different, which can reflect the target person.
  • the ability of the face name of the face in the face image is also different.
  • the title corresponding to a face image is “Star A, Star B is blocked by an international star”, and only the name of the person whose face image may be included can be obtained from the title;
  • "China's new generation popular popular star A" can find the name of the face corresponding to this picture. Therefore, the second sub-score can also reflect whether the name of the person is the name of the person's face in the target face picture.
  • the first name of the person name Name i is calculated as follows:
  • the first score of the person can be standardized:
  • weight p is the first sub-score and weight position is the second sub-score.
  • attributes of the text include, but are not limited to, a type of text, a location of the text, and/or a publisher of the text.
  • the type of the text for example, the person in the title is more likely to be the person name of the face in the target face picture, and the person in the body name is the person name of the face in the target face picture.
  • the possibility is lower; for the location of the text, the text is located on a well-known website, the person name is higher in the person's face image, and the text is on the non-known website.
  • the possibility of a person's face is lower; for publishers of text, it is more likely that the author in the text will publish the name of the person in the target face image, rather than the text published by the authority.
  • the person named is the person whose face name in the target face picture is less likely.
  • the relationship between the text and the name of the person in the text includes, but is not limited to, the position and/or number of occurrences of the person's name in the text in the text.
  • the position of the person's name in the text in the text for example, the person in the upper position in the text is more likely to be the person name of the face in the target face picture, and the position in the text is later. It is less likely that a person will be named as a person's face in the target face image.
  • another embodiment of the present invention provides a face picture name recognition device, and the face recognition module 520 includes:
  • the feature extraction module 521 is configured to acquire features of the collected face image and acquire features of the target face image.
  • the position of the face in the face image is detected, and the related features of the face are extracted, and the plurality of features form a multi-dimensional vector for comparison, for example, , 400-dimensional vector.
  • the feature comparison module 522 is configured to compare the feature of the collected face image with the feature of the target face image, and determine a similar face image according to the comparison result. In the present embodiment, a manner of effectively recognizing similar face pictures is provided by means of feature comparison.
  • Another embodiment of the present invention provides a face picture person name recognition apparatus, and the result score calculation module 533 calculates a product of a first point value of the same person name and a preset first weight value, and a number of the same person name The product of the binary value and the preset second weight, and the result score is calculated based on the obtained product.
  • the first weight and the second weight reflect the importance of the two methods of text and face recognition. If the first weight is set larger, the user is more concerned with the result of the text extraction, such as the second weight is set.
  • the user pays more attention to the result of face recognition; therefore, if the credibility of the text source is higher, the first weight can be set larger, if the algorithm of face recognition is better, then The second weight is set to be larger.
  • the technical solution of the embodiment is advantageous for the user to adjust the influence degree of the first score and the second score, thereby obtaining the most reasonable result score.
  • the result score is calculated for the person name Name i :
  • the person name is output as the person name of the face in the target face picture.
  • Another embodiment of the present invention provides a face picture person name recognition apparatus, and the text includes, but is not limited to, a title, a body, and/or a surrounding text of a target face picture in a document corresponding to the target face picture.
  • modules in the devices of the embodiments can be adaptively changed and placed in one or more devices different from the embodiment.
  • the modules or units or components of the embodiments may be combined into one module or unit or component, and further they may be divided into a plurality of sub-modules or sub-units or sub-components.
  • any combination of the features disclosed in the specification, including the accompanying claims, the abstract and the drawings, and any methods so disclosed, or All processes or units of the device are combined.
  • Each feature disclosed in this specification (including the accompanying claims, the abstract and the drawings) may be replaced by alternative features that provide the same, equivalent or similar purpose.
  • the various component embodiments of the present invention may be implemented in hardware, or in a software module running on one or more processors, or in a combination thereof.
  • a microprocessor or digital signal processor may be used in practice to implement some or all of the functionality of some or all of the components of the device or apparatus in accordance with embodiments of the present invention.
  • the invention can also be implemented as a device or device program (e.g., a computer program and a computer program product) for performing some or all of the methods described herein.
  • a program implementing the invention may be stored on a computer readable medium or may be in the form of one or more signals. Such signals may be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.
  • FIG. 8 illustrates a computing device that can implement a face picture person name recognition method in accordance with the present invention.
  • the computing device conventionally includes a processor 810 and a computer program product or computer readable medium in the form of a memory 820.
  • the memory 820 may be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), an EPROM, a hard disk, or a ROM.
  • the memory 820 has any of the above methods for performing
  • storage space 830 for program code may include various program code 831 for implementing various steps in the above methods, respectively.
  • the program code can be read from or written to one or more computer program products.
  • Such computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards or floppy disks.
  • Such a computer program product is typically a portable or fixed storage unit as described with reference to FIG.
  • the storage unit may have storage segments, storage spaces, and the like that are similar to the storage 820 in the computing device of FIG.
  • the program code can be compressed, for example, in an appropriate form.
  • the storage unit includes computer readable code 831', ie, code readable by a processor, such as 810, that when executed by a computing device causes the computing device to perform each of the methods described above step.

Abstract

一种人脸图片人名识别方法和装置,主要涉及计算机技术领域,主要目的在于准确识别出人脸图片中的人脸的人名。方法包括:从目标人脸图片对应的文本中提取人名;对目标人脸图片进行人脸识别,识别出目标人脸图片的相似人脸图片,并获取相似人脸图片中人脸的人名;根据文本中的人名,以及相似人脸图片中人脸的人名,确定目标人脸图片中人脸的人名。将通过人脸识别技术识别出的相似人脸的人名和从对应文本中得到的人名综合考虑,无疑能够更加准确地确定目标人脸图片中的人脸的人名。

Description

人脸图片人名识别方法和装置 技术领域
本发明涉及计算机技术领域,具体而言,涉及一种人脸图片人名识别方法和装置。
背景技术
目前,人脸识别技术正在逐渐发展,但是由于人脸图片往往存在光线、角度等方面的问题,容易造成难以准确识别出人脸对应人物的人名。
例如,某条新闻信息中报道了明星A的演唱会情况,并插入了具有明星A的脸部的照片,但基于现有的人脸识别技术进行识别后,得到结果为:照片中的人脸可能是明星A,也可能是明星B,且是明星B的可能性更大一些。此时,基于前述的人脸识别技术的识别结果来确定照片中人脸的人名,则非常容易出现错误。由于可见,仅基于人脸识别技术来确定人脸图片中的人脸的人名,准确率是较低的。
发明内容
鉴于上述问题,提出了本发明以便提供一种克服上述问题或者至少部分地解决上述问题的人脸图片人名识别方法和装置。
依据本发明的一个方面,提供了一种人脸图片人名识别方法,其包括:从目标人脸图片对应的文本中提取人名;对所述目标人脸图片进行人脸识别,识别出所述目标人脸图片的相似人脸图片,并获取所述相似人脸图片中人脸的人名;根据所述文本中的人名,以及所述相似人脸图片中人脸的人名,确定所述目标人脸图片中人脸的人名。
依据本发明的一个方面,还提供了一种人脸图片人名识别装置,其包括:人名提取模块,用于从目标人脸图片对应的文本中提取人名;人脸识别模块,用于对所述目标人脸图片进行人脸识别,识别出所述目标人脸图片的相似人脸图片,并获取所述相似人脸图片中人脸的人名;人名确定模块,用于根据所述文本中的人名,以及所述相似人脸图片中人脸的人名,确定所述目标人脸图片中人脸的人名。
依据本发明的一个方面,提供了一种计算机程序,其包括计算机可读代码,当所述计算机可读代码在计算设备上运行时,导致所述计算设备执行上述人脸图片人名识别方法。
依据本发明的一个方面,提供了一种计算机可读介质,其中存储了上述的计算机程序。
根据以上技术方案,可知本发明的人脸图片人名识别方法和装置至少具有以下优点:
在本发明的技术方案中,不但利用人脸识别技术来对目标人脸图片进行了识别,并且考虑到目标人脸图片对应的文本中所包含的人名,这是因为目标人脸图片对应的文本往往与目标人脸图片之间具有较紧密的关系,所以文本中涉及的人名包含了目标人脸图片中人脸的人名,而将通过人脸识别技术识别出的相似人脸的人名和从对应文本中得到的人名综合考虑,无疑能够更加准确地确定目标人脸图片中的人脸的人名。
上述说明仅是本发明技术方案的概述,为了能够更清楚了解本发明的技术手段,而可依照说明书的内容予以实施,并且为了让本发明的上述和其它目的、特征和优点能够更明显易懂,以下特举本发明的具体实施方式。
附图说明
通过阅读下文优选实施方式的详细描述,各种其他的优点和益处对于本领域普通技术人员将变得清楚明了。附图仅用于示出优选实施方式的目的,而并不认为是对本发明的限制。而且在整个附图中,用相同的参考符号表示相同的部件。在附图中:
图1示出了根据本发明的一个实施例的人脸图片人名识别方法的流程图;
图2示出了根据本发明的一个实施例的人脸图片人名识别方法的原理示意图;
图3示出了根据本发明的一个实施例的人脸图片人名识别方法的局部流程图;
图4示出了根据本发明的一个实施例的人脸图片人名识别方法的局部流程图;
图5示出了根据本发明的一个实施例的人脸图片人名识别装置的框图;
图6示出了根据本发明的一个实施例的人脸图片人名识别装置的模块框图;
图7示出了根据本发明的一个实施例的人脸图片人名识别装置的模块框图;
图8示意性地示出了用于执行根据本发明的人脸图片人名识别方法的计算设备的框图;以及
图9示意性地示出了用于保持或者携带实现根据本发明的人脸图片人名识别方法的程序代码的存储单元。
具体实施方式
下面将参照附图更详细地描述本公开的示例性实施例。虽然附图中显示了本公开的示例性实施例,然而应当理解,可以以各种形式实现本公开而不应被这里阐述的实施例所限制。相反,提供这些实施例是为了能够更透彻地理解本公开,并且能够将本公开的范围完整的传达给本领域的技术人员。
如图1所示,本发明的一个实施例提供了一种人脸图片人名识别方法,其包括:
步骤110,从目标人脸图片对应的文本中提取人名。在本实施例中,对于文本的类型不做限制,例如,当目标人脸图片位于一则新闻消息中时,则文本可以是该新闻消息的新闻内容;当目标人脸图片位于一篇科技文档中时,则文本可以是该目标人脸图片的上下文内容。在本实施例中,目标人脸图片通常与其对应文本之间存在紧密的联系,所以可以确定文本中很可能存在着目标人脸图片中的人脸的人名,例如,一则新闻消息中提供了一张明星的个人照片,而新闻内容报道了明星的演唱会情况,则新闻内容中提供了明星照片中明星的人名;一篇科技文档提供了一张科学家的照片,且照片下方的文本是对于该科学家的生平介绍,则照片下方文本中提供了科学家的人名。文本中所提取的人名数量为一个或多个。
步骤120,对目标人脸图片进行人脸识别,识别出目标人脸图片的相似人脸图片,并获取相似人脸图片中人脸的人名。在本实施例中,基于人脸识别技术识别出了目标人脸图片的相似人脸图片,且对所采用的人脸识别技术不进行限制,现有的任何人脸识别技术均可用于本实施例的技术方案。在本实施例中,可以在数据库中预存储大量已收集的人脸图片以及每张人脸图片中人脸的人名,则通过人脸识别技术可以从数据库中找到相似人脸图片,并能够确定相似人脸图片的人名。通过人脸识别技术得到的人名数量为一个或多个。
步骤130,根据文本中的人名,以及相似人脸图片中人脸的人名,确定目标人脸图片中人脸的人名。在本实施例中,由于综合考虑了人脸识别技术识别出的人名以及文本中包含的人名,所以能够更加准确地确定目标人脸图片中人脸的人名。在本实施例中,对于如何确定目标人脸图片中人脸的人名的方式不做限制,例如,对于一张新闻消息中的人脸图片,从该新闻消息的新闻内容中提取了“张三、李四”两个人名,通过人脸识别方式得到了目标人脸图片的两张相似人脸图片,并确定该两张相似人脸图片的名字分别为“李四、王五”,则同时被两种方式获取的“李四”的准确性比较较高,所以将“李四”确定为目标人脸图片中人脸的人名;而假如, 从该新闻消息的新闻内容中提取了“张三”一个人名,通过人脸识别方式得到了目标人脸图片的一张相似人脸图片,并确定该相似人脸图片的名字为“李四”,则以人脸识别技术得到的结果为准,将“李四”确定为目标人脸图片中人脸图片。
在本实施例的技术方案中,因为目标人脸图片对应的文本往往与目标人脸图片之间具有较紧密的关系,所以文本中涉及的人名包含了目标人脸图片中人脸的人名,而将通过人脸识别技术识别出的相似人脸的人名和从对应文本中得到的人名综合考虑,如示意图图2所示,无疑能够更加准确地确定目标人脸图片中的人脸的人名。
如图3所示,本发明的另一个实施例提供了一种人脸图片人名识别方法,步骤130具体包括:
步骤131,根据文本的属性和/或文本与文本中人名之间的关系,为文本中的人名计算第一分值。在本实施例中,文本的不同属性、文本与文本中人名之间的不同关系,能够反映人名为目标人脸图片中人脸的人名的可能性高低。例如,当文本的属性为文本的类型时,例如,一则新闻消息中提供了一张明星的个人照片,而新闻内容报道了明星的演唱会情况,则新闻内容中的人名有很大可能是明星的人名,则文本中每个人名的第一分值可设置为90分;一篇文学作品中提供了一张作者的照片,而其正文中包含的人名可能只是作品主角而非作者的人名,则文本中每个人名的第一分值可设置为10分。同理,当文本与文本中人名的关系为人名在文本中出现的次数时,例如,一则配有人脸图片的新闻消息中,人名1出现了9次,而人名2出现了一次,则容易理解人名1更可能是人脸图片中人脸的人名,所以人名1的第一分值可以设置为80分,而人名2的第一分值可设置为20分。本实施例中,对文本的属性、文本与文本中人名之间的关系不做限制。
步骤132,根据相似人脸图片与目标人脸图片的相似度,为相似人脸图片中人脸的人名计算第二分值。在本实施例中,容易理解,相似度更高的相似人脸图片中的人脸更可能与目标人脸图片中的人脸为同一张人脸,所以相似度较高的目标人脸图片应该具有更高的分值。假设返回的K张相似图片中,人名Namei出现了M次,对应的相似度分别为Similarityj,计算其第二分值:
Figure PCTCN2015091869-appb-000001
为便于计算,可将其进行标准化:
Figure PCTCN2015091869-appb-000002
需要说明的是,上述各个公式并不是实现本发明的唯一公式,仅作为实施例的一种实现方式。技术人员可以根据业务需要对公式做适当变形,依然落在本发明的范围之内,例如增添参数或倍数值等。
步骤133,根据相同人名具有的第一分值和第二分值计算结果分值,并根据结果分值的大小确定目标人脸图片中人脸的人名。在本实施例中,第一分值和第二分值都反映了所获得的人名为目标人脸图片中人脸的人名的可能性高低,所以综合第一分值和第二分值得到的结果分值,更能够反映每个人名是否为目标人脸图片中人脸的人名的可能性,则按结果分值的大小选择相应的人名作为目标人脸图片中人脸的人名,准确程度非常高。在本实施例中,对结果分值的计算方式不进行限制,例如,其可以是第一分值与第二分值的相加或相乘。例如,人名1的第一分值为80分、第二分值为60分,人名2的第一分值为90分、第二分值为20分,人名3只有第一分值为50分,结果分值的计算方式为第一分值和第二分值相加,则人名1的结果分值为140分、人名2的结果分值为110分,人名3的结果分值为50分,则选择人名1作为目标人脸图片的人名。
在本实施例的技术方案中,基于文本的属性、文本与文本中人名之间的关系得到的第一分值、基于相似人脸图片与目标人脸图片的相似度得到的第二分值,都反映了每个人名为目标人脸图片中人脸的人名的可能性高低,所以按照综合第一分值和第二分值得到的结果分值来选取人名作为目标人脸图片中人脸的人名,这种方式的准确性很高,尤其适用于通过文本提取、通过人脸识别技术得到了多个候选人名的情况。
本发明的另一个实施例提供了一种人脸图片人名识别方法,步骤131,具体包括:
根据文本的属性为文本中的人名计算第一子分值,根据文本与文本中人名的关系计算第二子分值,根据第一子分值和第二子分值计算第一分值。在本实施例中,同时考虑了文本的属性、文本与文本中人名的关系,所以计算得到的第一分值明显能够反映每个人名为目标人脸图片中人脸的人名的可能性高低。
例如,假设文本与文本中的人名的关系为人名在文本中的位置,容易理解人名 出现的位置会反映出该人名是否为目标人脸图片中人脸的人名的可能性高低,比如新闻消息中的最先出现的人名往往就是其中的人脸图片中人脸的人名,而较后位置出现的人名为目标人脸图片中人脸的人名的可能性较低,所以第一子分值也能够反映出该人名是否为目标人脸图片中人脸的人名的可能性高低。
例如,假设文本的属性为文本的类型,则一张人脸图片可能有多个文本,其类型可以能是人脸图片的标题、内容、环绕文本等;文本的类型不同,其能反映目标人脸图片中人脸的人名的能力也不同,比如一张人脸图片对应的标题是“明星A、明星B被封国际明星”,从标题仅能获得这张人脸图片可能包含的人名;再结合目标人脸图片对应的正文内容“中国新生代当红人气明星A”就可以找到这张图片对应的人脸的人名。所以,第二子分值也能够反映出该人名是否为目标人脸图片中人脸的人名的可能性高低。
由于同一人名可能出现在P个文本中,所以采用以下方式计算人名第一分值Namei
Figure PCTCN2015091869-appb-000003
假设共出现人名N个,可以将人名第一分值标准化:
Figure PCTCN2015091869-appb-000004
其中,weightp为第一子分值,weightposition为第二子分值。
需要说明的是,上述各个公式并不是实现本发明的唯一公式,仅作为实施例的一种实现方式。技术人员可以根据业务需要对公式做适当变形,依然落在本发明的范围之内,例如增添参数或倍数值等。
本发明的另一个实施例提供了一种人脸图片人名识别方法,文本的属性包括但不限于:文本的类型、文本的所在位置和/或文本的发布者。在本实施例中,对于文本的类型,例如,标题中的人名为目标人脸图片中人脸的人名的可能性较高,而正文中的人名为目标人脸图片中人脸的人名的可能性较较低;对于文本的所在位置,文本位于知名网站则人名为目标人脸图片中人脸的人名的可能性较高,而文本位于 非知名网站则人名为目标人脸图片中人脸的人名的可能性较较低;对于文本的发布者,较权威者发布文本中的人名为目标人脸图片中人脸的人名的可能性较高,而非权威者发布的文本的人名为目标人脸图片中人脸的人名的可能性较较低。
文本与文本中的人名的关系包括但不限于:文本中的人名在文本中的位置和/或出现次数。在本实施例中,对于文本中的人名在文本中的位置,例如,文本中较前位置的人名为目标人脸图片中人脸的人名的可能性较高,而文本中较后位置的人名为目标人脸图片中人脸的人名的可能性较较低。
如图4所示,本发明的另一个实施例提供了一种人脸图片人名识别方法,步骤120具体包括:
步骤121,获取已收集的人脸图片的特征,以及获取目标人脸图片的特征。在本实施例中,基于人脸特征提取,并通过图像处理算法,检测出人脸图片中人脸的位置,并提取人脸的相关特征,多个特征形成多维向量以用于进行比较,例如,400维的向量。
步骤122,将已收集的人脸图片的特征与目标人脸图片的特征进行比较,并根据比较结果确定相似人脸图片。在本实施例中,通过特征比较的方式,提供了一种有效地识别相似人脸图片的方式。
本发明的另一个实施例提供了一种人脸图片人名识别方法,步骤130具体包括:
计算相同人名具有的第一分值与预设的第一权值的乘积,以及相同人名具有的第二分值与预设的第二权值的乘积,并根据得到的乘积计算结果分值。
在本实施例中,第一权值和第二权值反映了正文和人脸识别两种方式的重要程度,如第一权值设置得较大,则说明用户较看重文本提取的结果,如第二权值设置得较大,则说明用户较看重人脸识别的结果;因此,如果文本来源的可信度较高,则可以将第一权值设置得较大,如果人脸识别的算法较优,则可以将第二权值设置得较大,本实施例的技术方案有利于用户调整第一分值和第二分值的影响程度,从而获得最合理的结果分值。
基于前述实施例的公式,对于人名Namei计算其结果分值:
Figure PCTCN2015091869-appb-000005
然后找出结果分值最高的人名:
Name=MAX(Namei)
如果所选人名的得分大于指定阈值Threshold,则将此人名作为目标人脸图片中人脸的人名进行输出。
需要说明的是,上述各个公式并不是实现本发明的唯一公式,仅作为实施例的一种实现方式。技术人员可以根据业务需要对公式做适当变形,依然落在本发明的范围之内,例如增添参数或倍数值等。
本发明的另一个实施例提供了一种人脸图片人名识别方法,文本包括但不限于目标人脸图片对应的文档中的标题、正文和/或目标人脸图片的环绕文本。
如图5所示,本发明的另一个实施例提供了一种人脸图片人名识别装置,其包括:
人名提取模块510,用于从目标人脸图片对应的文本中提取人名。在本实施例中,对于文本的类型不做限制,例如,当目标人脸图片位于一则新闻消息中时,则文本可以是该新闻消息的新闻内容;当目标人脸图片位于一篇科技文档中时,则文本可以是该目标人脸图片的上下文内容。在本实施例中,目标人脸图片通常与其对应文本之间存在紧密的联系,所以可以确定文本中很可能存在着目标人脸图片中的人脸的人名,例如,一则新闻消息中提供了一张明星的个人照片,而新闻内容报道了明星的演唱会情况,则新闻内容中提供了明星照片中明星的人名;一篇科技文档提供了一张科学家的照片,且照片下方的文本是对于该科学家的生平介绍,则照片下方文本中提供了科学家的人名。文本中所提取的人名数量为一个或多个。
人脸识别模块520,用于对目标人脸图片进行人脸识别,识别出目标人脸图片的相似人脸图片,并获取相似人脸图片中人脸的人名。在本实施例中,基于人脸识别技术识别出了目标人脸图片的相似人脸图片,且对所采用的人脸识别技术不进行限制,现有的任何人脸识别技术均可用于本实施例的技术方案。在本实施例中,可以在数据库中预存储大量已收集的人脸图片以及每张人脸图片中人脸的人名,则通过人脸识别技术可以从数据库中找到相似人脸图片,并能够确定相似人脸图片的人名。通过人脸识别技术得到的人名数量为一个或多个。
人名确定模块530,用于根据文本中的人名,以及相似人脸图片中人脸的人名, 确定目标人脸图片中人脸的人名。在本实施例中,由于综合考虑了人脸识别技术识别出的人名以及文本中包含的人名,所以能够更加准确地确定目标人脸图片中人脸的人名。在本实施例中,对于如何确定目标人脸图片中人脸的人名的方式不做限制,例如,对于一张新闻消息中的人脸图片,从该新闻消息的新闻内容中提取了“张三、李四”两个人名,通过人脸识别方式得到了目标人脸图片的两张相似人脸图片,并确定该两张相似人脸图片的名字分别为“李四、王五”,则同时被两种方式获取的“李四”的准确性比较较高,所以将“李四”确定为目标人脸图片中人脸的人名;而假如,从该新闻消息的新闻内容中提取了“张三”一个人名,通过人脸识别方式得到了目标人脸图片的一张相似人脸图片,并确定该相似人脸图片的名字为“李四”,则以人脸识别技术得到的结果为准,将“李四”确定为目标人脸图片中人脸图片。
在本实施例的技术方案中,因为目标人脸图片对应的文本往往与目标人脸图片之间具有较紧密的关系,所以文本中涉及的人名包含了目标人脸图片中人脸的人名,而将通过人脸识别技术识别出的相似人脸的人名和从对应文本中得到的人名综合考虑,如示意图图2所示,无疑能够更加准确地确定目标人脸图片中的人脸的人名。
如图6所示,本发明的另一个实施例提供了一种人脸图片人名识别装置,人名确定模块530包括:
第一分值计算模块531,用于根据文本的属性和/或文本与文本中人名之间的关系,为文本中的人名计算第一分值。在本实施例中,文本的不同属性、文本与文本中人名之间的不同关系,能够反映人名为目标人脸图片中人脸的人名的可能性高低。例如,当文本的属性为文本的类型时,例如,一则新闻消息中提供了一张明星的个人照片,而新闻内容报道了明星的演唱会情况,则新闻内容中的人名有很大可能是明星的人名,则文本中每个人名的第一分值可设置为90分;一篇文学作品中提供了一张作者的照片,而其正文中包含的人名可能只是作品主角而非作者的人名,则文本中每个人名的第一分值可设置为10分。同理,当文本与文本中人名的关系为人名在文本中出现的次数时,例如,一则配有人脸图片的新闻消息中,人名1出现了9次,而人名2出现了一次,则容易理解人名1更可能是人脸图片中人脸的人名,所以人名1的第一分值可以设置为80分,而人名2的第一分值可设置为20分。本实施例中,对文本的属性、文本与文本中人名之间的关系不做限制。
第二分值计算模块532,用于根据相似人脸图片与目标人脸图片的相似度,为相 似人脸图片中人脸的人名计算第二分值。在本实施例中,容易理解,相似度更高的相似人脸图片中的人脸更可能与目标人脸图片中的人脸为同一张人脸,所以相似度较高的目标人脸图片应该具有更高的分值。假设返回的K张相似图片中,人名Namei出现了M次,对应的相似度分别为Similarityj,计算其第二分值:
Figure PCTCN2015091869-appb-000006
为便于计算,可将其进行标准化:
Figure PCTCN2015091869-appb-000007
需要说明的是,上述各个公式并不是实现本发明的唯一公式,仅作为实施例的一种实现方式。技术人员可以根据业务需要对公式做适当变形,依然落在本发明的范围之内,例如增添参数或倍数值等。
结果分值计算模块533,用于根据相同人名具有的第一分值和第二分值计算结果分值,并根据结果分值的大小确定目标人脸图片中人脸的人名。在本实施例中,第一分值和第二分值都反映了所获得的人名为目标人脸图片中人脸的人名的可能性高低,所以综合第一分值和第二分值得到的结果分值,更能够反映每个人名是否为目标人脸图片中人脸的人名的可能性,则按结果分值的大小选择相应的人名作为目标人脸图片中人脸的人名,准确程度非常高。在本实施例中,对结果分值的计算方式不进行限制,例如,其可以是第一分值与第二分值的相加或相乘。例如,人名1的第一分值为80分、第二分值为60分,人名2的第一分值为90分、第二分值为20分,人名3只有第一分值为50分,结果分值的计算方式为第一分值和第二分值相加,则人名1的结果分值为140分、人名2的结果分值为110分,人名3的结果分值为50分,则选择人名1作为目标人脸图片的人名。
在本实施例的技术方案中,基于文本的属性、文本与文本中人名之间的关系得到的第一分值、基于相似人脸图片与目标人脸图片的相似度得到的第二分值,都反映了每个人名为目标人脸图片中人脸的人名的可能性高低,所以按照综合第一分值和第二分值得到的结果分值来选取人名作为目标人脸图片中人脸的人名,这种方式 的准确性很高,尤其适用于通过文本提取、通过人脸识别技术得到了多个候选人名的情况。
本发明的另一个实施例提供了一种人脸图片人名识别装置,第一分值计算模块531根据文本的属性为文本中的人名计算第一子分值,根据文本与文本中人名的关系计算第二子分值,根据第一子分值和第二子分值计算第一分值。在本实施例中,同时考虑了文本的属性、文本与文本中人名的关系,所以计算得到的第一分值明显能够反映每个人名为目标人脸图片中人脸的人名的可能性高低。
例如,假设文本与文本中的人名的关系为人名在文本中的位置,容易理解人名出现的位置会反映出该人名是否为目标人脸图片中人脸的人名的可能性高低,比如新闻消息中的最先出现的人名往往就是其中的人脸图片中人脸的人名,而较后位置出现的人名为目标人脸图片中人脸的人名的可能性较低,所以第一子分值也能够反映出该人名是否为目标人脸图片中人脸的人名的可能性高低。
例如,假设文本的属性为文本的类型,则一张人脸图片可能有多个文本,其类型可以能是人脸图片的标题、内容、环绕文本等;文本的类型不同,其能反映目标人脸图片中人脸的人名的能力也不同,比如一张人脸图片对应的标题是“明星A、明星B被封国际明星”,从标题仅能获得这张人脸图片可能包含的人名;再结合目标人脸图片对应的正文内容“中国新生代当红人气明星A”就可以找到这张图片对应的人脸的人名。所以,第二子分值也能够反映出该人名是否为目标人脸图片中人脸的人名的可能性高低。
由于同一人名可能出现在P个文本中,所以采用以下方式计算人名第一分值Namei
Figure PCTCN2015091869-appb-000008
假设共出现人名N个,可以将人名第一分值标准化:
Figure PCTCN2015091869-appb-000009
其中,weightp为第一子分值,weightposition为第二子分值。
需要说明的是,上述各个公式并不是实现本发明的唯一公式,仅作为实施例的一种实现方式。技术人员可以根据业务需要对公式做适当变形,依然落在本发明的范围之内,例如增添参数或倍数值等。
本发明的另一个实施例提供了一种人脸图片人名识别装置,文本的属性包括但不限于:文本的类型、文本的所在位置和/或文本的发布者。在本实施例中,对于文本的类型,例如,标题中的人名为目标人脸图片中人脸的人名的可能性较高,而正文中的人名为目标人脸图片中人脸的人名的可能性较较低;对于文本的所在位置,文本位于知名网站则人名为目标人脸图片中人脸的人名的可能性较高,而文本位于非知名网站则人名为目标人脸图片中人脸的人名的可能性较较低;对于文本的发布者,较权威者发布文本中的人名为目标人脸图片中人脸的人名的可能性较高,而非权威者发布的文本的人名为目标人脸图片中人脸的人名的可能性较较低。
文本与文本中的人名的关系包括但不限于:文本中的人名在文本中的位置和/或出现次数。在本实施例中,对于文本中的人名在文本中的位置,例如,文本中较前位置的人名为目标人脸图片中人脸的人名的可能性较高,而文本中较后位置的人名为目标人脸图片中人脸的人名的可能性较较低。
如图7所示,本发明的另一个实施例提供了一种人脸图片人名识别装置,人脸识别模块520包括:
特征提取模块521,用于获取已收集的人脸图片的特征,以及获取目标人脸图片的特征。在本实施例中,基于人脸特征提取,并通过图像处理算法,检测出人脸图片中人脸的位置,并提取人脸的相关特征,多个特征形成多维向量以用于进行比较,例如,400维的向量。
特征比较模块522,用于将已收集的人脸图片的特征与目标人脸图片的特征进行比较,并根据比较结果确定相似人脸图片。在本实施例中,通过特征比较的方式,提供了一种有效地识别相似人脸图片的方式。
本发明的另一个实施例提供了一种人脸图片人名识别装置,结果分值计算模块533计算相同人名具有的第一分值与预设的第一权值的乘积,以及相同人名具有的第二分值与预设的第二权值的乘积,并根据得到的乘积计算结果分值。在本实施例中, 第一权值和第二权值反映了正文和人脸识别两种方式的重要程度,如第一权值设置得较大,则说明用户较看重文本提取的结果,如第二权值设置得较大,则说明用户较看重人脸识别的结果;因此,如果文本来源的可信度较高,则可以将第一权值设置得较大,如果人脸识别的算法较优,则可以将第二权值设置得较大,本实施例的技术方案有利于用户调整第一分值和第二分值的影响程度,从而获得最合理的结果分值。
基于前述实施例的公式,对于人名Namei计算其结果分值:
然后找出结果分值最高的人名:
Name=MAX(Namei)
如果所选人名的得分大于指定阈值Threshold,则将此人名作为目标人脸图片中人脸的人名进行输出。
需要说明的是,上述各个公式并不是实现本发明的唯一公式,仅作为实施例的一种实现方式。技术人员可以根据业务需要对公式做适当变形,依然落在本发明的范围之内,例如增添参数或倍数值等。
本发明的另一个实施例提供了一种人脸图片人名识别装置,文本包括但不限于目标人脸图片对应的文档中的标题、正文和/或目标人脸图片的环绕文本。
在此提供的算法和显示不与任何特定计算机、虚拟系统或者其它设备固有相关。各种通用系统也可以与基于在此的示教一起使用。根据上面的描述,构造这类系统所要求的结构是显而易见的。此外,本发明也不针对任何特定编程语言。应当明白,可以利用各种编程语言实现在此描述的本发明的内容,并且上面对特定语言所做的描述是为了披露本发明的最佳实施方式。
在此处所提供的说明书中,说明了大量具体细节。然而,能够理解,本发明的实施例可以在没有这些具体细节的情况下实践。在一些实例中,并未详细示出公知的方法、结构和技术,以便不模糊对本说明书的理解。
类似地,应当理解,为了精简本公开并帮助理解各个发明方面中的一个或多个,在上面对本发明的示例性实施例的描述中,本发明的各个特征有时被一起分组到单 个实施例、图、或者对其的描述中。然而,并不应将该公开的方法解释成反映如下意图:即所要求保护的本发明要求比在每个权利要求中所明确记载的特征更多的特征。更确切地说,如下面的权利要求书所反映的那样,发明方面在于少于前面公开的单个实施例的所有特征。因此,遵循具体实施方式的权利要求书由此明确地并入该具体实施方式,其中每个权利要求本身都作为本发明的单独实施例。
本领域那些技术人员可以理解,可以对实施例中的设备中的模块进行自适应性地改变并且把它们设置在与该实施例不同的一个或多个设备中。可以把实施例中的模块或单元或组件组合成一个模块或单元或组件,以及此外可以把它们分成多个子模块或子单元或子组件。除了这样的特征和/或过程或者单元中的至少一些是相互排斥之外,可以采用任何组合对本说明书(包括伴随的权利要求、摘要和附图)中公开的所有特征以及如此公开的任何方法或者设备的所有过程或单元进行组合。除非另外明确陈述,本说明书(包括伴随的权利要求、摘要和附图)中公开的每个特征可以由提供相同、等同或相似目的的替代特征来代替。
此外,本领域的技术人员能够理解,尽管在此所述的一些实施例包括其它实施例中所包括的某些特征而不是其它特征,但是不同实施例的特征的组合意味着处于本发明的范围之内并且形成不同的实施例。例如,在下面的权利要求书中,所要求保护的实施例的任意之一都可以以任意的组合方式来使用。
本发明的各个部件实施例可以以硬件实现,或者以在一个或者多个处理器上运行的软件模块实现,或者以它们的组合实现。本领域的技术人员应当理解,可以在实践中使用微处理器或者数字信号处理器(DSP)来实现根据本发明实施例的设备或装置中的一些或者全部部件的一些或者全部功能。本发明还可以实现为用于执行这里所描述的方法的一部分或者全部的设备或者装置程序(例如,计算机程序和计算机程序产品)。这样的实现本发明的程序可以存储在计算机可读介质上,或者可以具有一个或者多个信号的形式。这样的信号可以从因特网网站上下载得到,或者在载体信号上提供,或者以任何其他形式提供。
例如,图8示出了可以实现根据本发明的人脸图片人名识别方法的计算设备。该计算设备传统上包括处理器810和以存储器820形式的计算机程序产品或者计算机可读介质。存储器820可以是诸如闪存、EEPROM(电可擦除可编程只读存储器)、EPROM、硬盘或者ROM之类的电子存储器。存储器820具有用于执行上述方法中的任 何方法步骤的程序代码831的存储空间830。例如,用于程序代码的存储空间830可以包括分别用于实现上面的方法中的各种步骤的各个程序代码831。这些程序代码可以从一个或者多个计算机程序产品中读出或者写入到这一个或者多个计算机程序产品中。这些计算机程序产品包括诸如硬盘,紧致盘(CD)、存储卡或者软盘之类的程序代码载体。这样的计算机程序产品通常为如参考图9所述的便携式或者固定存储单元。该存储单元可以具有与图8的计算设备中的存储器820类似布置的存储段、存储空间等。程序代码可以例如以适当形式进行压缩。通常,存储单元包括计算机可读代码831’,即可以由例如诸如810之类的处理器读取的代码,这些代码当由计算设备运行时,导致该计算设备执行上面所描述的方法中的各个步骤。
本文中所称的“一个实施例”、“实施例”或者“一个或者多个实施例”意味着,结合实施例描述的特定特征、结构或者特性包括在本发明的至少一个实施例中。此外,请注意,这里“在一个实施例中”的词语例子不一定全指同一个实施例。
应该注意的是上述实施例对本发明进行说明而不是对本发明进行限制,并且本领域技术人员在不脱离所附权利要求的范围的情况下可设计出替换实施例。在权利要求中,不应将位于括号之间的任何参考符号构造成对权利要求的限制。单词“包含”不排除存在未列在权利要求中的元件或步骤。位于元件之前的单词“一”或“一个”不排除存在多个这样的元件。本发明可以借助于包括有若干不同元件的硬件以及借助于适当编程的计算机来实现。在列举了若干装置的单元权利要求中,这些装置中的若干个可以是通过同一个硬件项来具体体现。单词第一、第二、以及第三等的使用不表示任何顺序。可将这些单词解释为名称。
至此,本领域技术人员应认识到,虽然本文已详尽示出和描述了本发明的多个示例性实施例,但是,在不脱离本发明精神和范围的情况下,仍可根据本发明公开的内容直接确定或推导出符合本发明原理的许多其他变型或修改。因此,本发明的范围应被理解和认定为覆盖了所有这些其他变型或修改。
此外,还应当注意,本说明书中使用的语言主要是为了可读性和教导的目的而选择的,而不是为了解释或者限定本发明的主题而选择的。因此,在不偏离所附权利要求书的范围和精神的情况下,对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。对于本发明的范围,对本发明所做的公开是说明性的,而非限制性的,本发明的范围由所附权利要求书限定。

Claims (16)

  1. 一种人脸图片人名识别方法,其包括:
    从目标人脸图片对应的文本中提取人名;
    对所述目标人脸图片进行人脸识别,识别出所述目标人脸图片的相似人脸图片,并获取所述相似人脸图片中人脸的人名;
    根据所述文本中的人名,以及所述相似人脸图片中人脸的人名,确定所述目标人脸图片中人脸的人名。
  2. 根据权利要求1所述的方法,其中,根据所述文本中的人名,以及所述相似人脸图片中人脸的人名,确定所述目标人脸图片中人脸的人名,具体包括:
    根据所述文本的属性和/或所述文本与所述文本中人名之间的关系,为所述文本中的人名计算第一分值;
    根据所述相似人脸图片与所述目标人脸图片的相似度,为所述相似人脸图片中人脸的人名计算第二分值;
    根据相同人名具有的第一分值和第二分值计算结果分值,并根据结果分值的大小确定所述目标人脸图片中人脸的人名。
  3. 根据权利要求1-2任一项所述的方法,其中,根据所述文本的属性和/或所述文本与所述文本中人名之间的关系,为所述文本中的人名计算第一分值,具体包括:
    根据所述文本的属性为所述文本中的人名计算第一子分值,根据所述文本与所述文本中人名的关系计算第二子分值,根据所述第一子分值和所述第二子分值计算所述第一分值。
  4. 根据权利要求1-3任一项所述的方法,其中,
    所述文本的属性包括:所述文本的类型、所述文本的所在位置和/或所述文本的发布者;
    所述文本与所述文本中的人名的关系包括:所述文本中的人名在所述文本中的位置和/或出现次数。
  5. 根据权利要求1-4任一项所述的方法,其中,对所述目标人脸图片进行人脸识别,识别出所述目标人脸图片的相似人脸图片,具体包括:
    获取已收集的人脸图片的特征,以及获取所述目标人脸图片的特征;
    将所述已收集的人脸图片的特征与所述目标人脸图片的特征进行比较,并根据比较结果确定所述相似人脸图片。
  6. 根据权利要求1-5任一项所述的方法,其中,根据相同人名具有的第一分值和第二分值计算结果分值,具体包括:
    计算所述相同人名具有的第一分值与预设的第一权值的乘积,以及所述相同人名具有的第二分值与预设的第二权值的乘积,并根据得到的乘积计算所述结果分值。
  7. 根据权利要求1至6中任一项所述的方法,其中,
    所述文本包括所述目标人脸图片对应的文档中的标题、正文和/或所述目标人脸图片的环绕文本。
  8. 一种人脸图片人名识别装置,其包括:
    人名提取模块,用于从目标人脸图片对应的文本中提取人名;
    人脸识别模块,用于对所述目标人脸图片进行人脸识别,识别出所述目标人脸图片的相似人脸图片,并获取所述相似人脸图片中人脸的人名;
    人名确定模块,用于根据所述文本中的人名,以及所述相似人脸图片中人脸的人名,确定所述目标人脸图片中人脸的人名。
  9. 根据权利要求8所述的装置,其中,所述人名确定模块包括:
    第一分值计算模块,用于根据所述文本的属性和/或所述文本与所述文本中人名之间的关系,为所述文本中的人名计算第一分值;
    第二分值计算模块,用于根据所述相似人脸图片与所述目标人脸图片的相似度,为所述相似人脸图片中人脸的人名计算第二分值;
    结果分值计算模块,用于根据相同人名具有的第一分值和第二分值计算结果分值,并根据结果分值的大小确定所述目标人脸图片中人脸的人名。
  10. 根据权利要求8-9任一项所述的装置,其中,
    所述第一分值计算模块根据所述文本的属性为所述文本中的人名计算第一子分值,根据所述文本与所述文本中人名的关系计算第二子分值,根据所述第一子分值和所述第二子分值计算所述第一分值。
  11. 根据权利要求9所述的装置,其中,
    所述文本的属性包括:所述文本的类型、所述文本的所在位置和/或所述文本的发布者;
    所述文本与所述文本中的人名的关系包括:所述文本中的人名在所述文本中的位置和/或出现次数。
  12. 根据权利要求8所述的装置,其中,所述人脸识别模块包括:
    特征提取模块,用于获取已收集的人脸图片的特征,以及获取所述目标人脸图片的特征;
    特征比较模块,用于将所述已收集的人脸图片的特征与所述目标人脸图片的特征进行比较,并根据比较结果确定所述相似人脸图片。
  13. 根据权利要求9所述的装置,其中,
    所述结果分值计算模块计算所述相同人名具有的第一分值与预设的第一权值的乘积,以及所述相同人名具有的第二分值与预设的第二权值的乘积,并根据得到的乘积计算所述结果分值。
  14. 根据权利要求8至13中任一项所述的装置,其中,
    所述文本包括所述目标人脸图片对应的文档中的标题、正文和/或所述目标人脸图片的环绕文本。
  15. 一种计算机程序,包括计算机可读代码,当所述计算机可读代码在计算设备上运行时,导致所述计算设备执行根据权利要求1-7中的任一个所述的人脸图片人名识别方法。
  16. 一种计算机可读介质,其中存储了如权利要求15所述的计算机程序。
PCT/CN2015/091869 2014-10-13 2015-10-13 人脸图片人名识别方法和装置 WO2016058520A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201410539153.3 2014-10-13
CN201410539153.3A CN104281842A (zh) 2014-10-13 2014-10-13 人脸图片人名识别方法和装置

Publications (1)

Publication Number Publication Date
WO2016058520A1 true WO2016058520A1 (zh) 2016-04-21

Family

ID=52256702

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/091869 WO2016058520A1 (zh) 2014-10-13 2015-10-13 人脸图片人名识别方法和装置

Country Status (2)

Country Link
CN (1) CN104281842A (zh)
WO (1) WO2016058520A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10997609B1 (en) 2020-01-06 2021-05-04 Alipay Labs (singapore) Pte. Ltd. Biometric based user identity verification

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104281842A (zh) * 2014-10-13 2015-01-14 北京奇虎科技有限公司 人脸图片人名识别方法和装置
CN107943892B (zh) * 2017-11-16 2021-12-21 海信集团有限公司 一种视频中主要角色名称的确定方法及装置
CN108882033B (zh) * 2018-07-19 2021-12-14 上海影谱科技有限公司 一种基于视频语音的人物识别方法、装置、设备和介质
CN113869281A (zh) * 2018-07-19 2021-12-31 北京影谱科技股份有限公司 一种人物识别方法、装置、设备和介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101739428A (zh) * 2008-11-10 2010-06-16 中国科学院计算技术研究所 一种对多媒体创建索引的方法
US20120114197A1 (en) * 2010-11-09 2012-05-10 Microsoft Corporation Building a person profile database
JP2013235329A (ja) * 2012-05-07 2013-11-21 Taiwan Colour & Imaging Technology Corp 顔識別監視管理方法
CN104091164A (zh) * 2014-07-28 2014-10-08 北京奇虎科技有限公司 人脸图片人名识别方法和系统
CN104281842A (zh) * 2014-10-13 2015-01-14 北京奇虎科技有限公司 人脸图片人名识别方法和装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120117051A1 (en) * 2010-11-05 2012-05-10 Microsoft Corporation Multi-modal approach to search query input
US20130216107A1 (en) * 2012-02-20 2013-08-22 Chih-Hsung Huang Method of surveillance by face recognition
CN103793697B (zh) * 2014-02-17 2018-05-01 北京旷视科技有限公司 一种人脸图像的身份标注方法及人脸身份识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101739428A (zh) * 2008-11-10 2010-06-16 中国科学院计算技术研究所 一种对多媒体创建索引的方法
US20120114197A1 (en) * 2010-11-09 2012-05-10 Microsoft Corporation Building a person profile database
JP2013235329A (ja) * 2012-05-07 2013-11-21 Taiwan Colour & Imaging Technology Corp 顔識別監視管理方法
CN104091164A (zh) * 2014-07-28 2014-10-08 北京奇虎科技有限公司 人脸图片人名识别方法和系统
CN104281842A (zh) * 2014-10-13 2015-01-14 北京奇虎科技有限公司 人脸图片人名识别方法和装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10997609B1 (en) 2020-01-06 2021-05-04 Alipay Labs (singapore) Pte. Ltd. Biometric based user identity verification

Also Published As

Publication number Publication date
CN104281842A (zh) 2015-01-14

Similar Documents

Publication Publication Date Title
WO2016058520A1 (zh) 人脸图片人名识别方法和装置
WO2016015621A1 (zh) 人脸图片人名识别方法和系统
US9436682B2 (en) Techniques for machine language translation of text from an image based on non-textual context information from the image
US8649602B2 (en) Systems and methods for tagging photos
US11461386B2 (en) Visual recognition using user tap locations
WO2016180268A1 (zh) 一种文本聚合方法及装置
US10810374B2 (en) Matching a query to a set of sentences using a multidimensional relevancy determination
US10650274B2 (en) Image clustering method, image clustering system, and image clustering server
WO2015184992A1 (zh) 一种识别重复图片的方法、图片搜索去重方法及其装置
WO2015196964A1 (zh) 搜索匹配图片的方法、图片搜索方法及装置
WO2016101766A1 (zh) 相似人脸图片获取和人脸图片信息获取方法和装置
WO2015188719A1 (zh) 结构化数据与图片的关联方法与关联装置
WO2019028990A1 (zh) 代码元素的命名方法、装置、电子设备及介质
CN106156794B (zh) 基于文字风格识别的文字识别方法及装置
WO2015131528A1 (zh) 确定给定文本的主题分布的方法和装置
WO2016155626A1 (zh) 实现搜索提示的装置、系统及方法
TW201421267A (zh) 搜索系統及方法
CN111695405B (zh) 一种狗脸特征点的检测方法、装置、系统及存储介质
WO2016101737A1 (zh) 搜索查询方法和装置
US20210019553A1 (en) Information processing apparatus, control method, and program
JP5845988B2 (ja) 画像処理システム、画像処理方法、サーバおよびプログラム
WO2017107651A1 (zh) 确定新闻之间相关性、多新闻之间相关性计算方法和装置
TW201921268A (zh) 用於分割文本的系統和方法
KR101698280B1 (ko) 태그에 대한 웹 페이지 검색 장치 및 방법
WO2015070678A1 (zh) 一种图像识别方法、挖掘图像主体信息方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 15851094

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 15851094

Country of ref document: EP

Kind code of ref document: A1