WO2013008427A1 - 画像評価装置、画像評価方法、プログラム、および集積回路 - Google Patents
画像評価装置、画像評価方法、プログラム、および集積回路 Download PDFInfo
- Publication number
- WO2013008427A1 WO2013008427A1 PCT/JP2012/004368 JP2012004368W WO2013008427A1 WO 2013008427 A1 WO2013008427 A1 WO 2013008427A1 JP 2012004368 W JP2012004368 W JP 2012004368W WO 2013008427 A1 WO2013008427 A1 WO 2013008427A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- clothes
- type information
- type
- person
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/80—Recognising image objects characterised by unique random patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Definitions
- the present invention relates to an image evaluation apparatus for evaluating an image using clothes information.
- Patent Document 1 proposes a method of evaluating a photographed event based on a recognition result of clothes of a person shown in an image and classifying the image according to the evaluation result.
- a clothes area an area considered to be clothes (hereinafter referred to as a clothes area) from the image and extract an image feature in the clothes area.
- the image feature amount include the color content ratio in the clothing area, and the amount of change in luminance between adjacent pixels.
- the present invention has been made in view of such problems, and an object of the present invention is to provide an image evaluation apparatus capable of correctly evaluating an image even if the above-described misrecognition occurs.
- the image evaluation apparatus comprises first specifying means for specifying first type information indicating the type of clothes worn by the person for each person shown in the image, and a plurality of pieces belonging to a predetermined image group.
- Second specifying means for specifying second type information indicating the type of clothes characterizing the predetermined image group based on the appearance frequency of each type of the first type information specified from the image; and the second type information
- an evaluation unit that evaluates an event in which a plurality of images belonging to the predetermined image group are captured.
- Example of image management information table Example of person management information table An example of three images and face and clothes areas detected from those images An example of two images and face and clothes regions detected from those images Flowchart of Processing of Image Evaluation Device 100 According to Embodiment 1 Flow chart of processing of image event evaluation unit 107 in the first embodiment An example of a table of third type information and the number of images An example of a table showing correspondence between clothes and events An example of a table showing event evaluation results of image groups
- Functional block diagram of the image evaluation device 1100 in the second embodiment An example of a similarity information table between clothes of persons Flow chart of processing of the image evaluation device 1100 in the second embodiment Flow chart of processing of image event evaluation unit 1102 according to the second embodiment
- Functional block diagram of the image evaluation apparatus 1500 in the third embodiment An example of a table showing clusters by face Flow chart of processing of image evaluation device 1500 in the third embodiment Flow chart of processing of image event evaluation unit 1503 in the third embodiment An example of a table showing faces for each cluster An example
- FIG. 1 is a functional block diagram of an image evaluation apparatus 100 according to the first embodiment. As shown in FIG. 1, the image evaluation device 100 is connected to the imaging device 120 and the display device 130.
- the image evaluation device 100 acquires an image group from the imaging device 120, evaluates the acquired image group, and outputs the image group to the display device 130 according to the evaluation result.
- the imaging device 120 captures an image and stores the captured image.
- the imaging device 120 is configured of, for example, a digital camera or the like, and is connected to the image evaluation device 100 via a USB (Universal Serial Bus) cable or the like.
- USB Universal Serial Bus
- the display device 130 displays an image such as an image output from the image evaluation device 100.
- the display device 130 is configured of, for example, a digital television, and is connected to the image evaluation device 100 via an HDMI (High Definition Multimedia Interface) cable or the like.
- HDMI High Definition Multimedia Interface
- the image evaluation apparatus 100 includes an image information acquisition unit 110, an image event evaluation unit 107, and a storage unit 108.
- the image information acquisition unit 110 further includes an image acquisition unit 101, an image group generation unit 102, a face detection unit 103, a clothes detection unit 104, a clothes feature extraction unit 105, and a clothes recognition unit 106.
- the image acquisition unit 101 collectively acquires the image group accumulated in the imaging device 120, and assigns a unique image ID to each of the acquired images.
- the image acquisition unit 101 registers the image ID attached to each image in the image management information table 201 shown in FIG.
- the image group generation unit 102 classifies the image group acquired by the image acquisition unit 101 into a plurality of image groups.
- the image group generation method takes, for example, images taken on the same day as one image group, and assigns a unique image group ID to each image group.
- the image group generation unit 102 acquires the shooting date and time from Exchangeable Image File Format (EXIF) information attached to the image, and uses it for classification of the image group.
- EXIF Exchangeable Image File Format
- the image group generation unit 102 registers the shooting date and time and the image group ID obtained as described above in the image management information table 201.
- the face detection unit 103 detects a square area (coordinates in the image) in which a human face appears from each image of the image group acquired by the image acquisition unit 101, and assigns a unique face ID to each of the detected faces. Then, the face detection unit 103 registers the face ID in the image management information table 201 and the person management information table 301 shown in FIG. 3 held by the storage unit 108.
- the face area is detected, for example, by matching using a face learning dictionary composed of face images prepared in advance.
- the clothes detection unit 104 detects, based on the coordinates of the face area detected by the face detection unit 103, an area in which the clothes of the person of the face are shown.
- the clothes detection unit 104 detects a clothes area obtained by calculating the ratio of the face, the neck and the upper body from the position and size of the face area, and manages the detected clothes area in association with the face ID.
- the clothes area is 0.2 down from the lower end of the face area.
- the area is 2.8 x 2.0 wide apart.
- the clothing area is an area of 280 pixels vertically by 200 pixels horizontally, which is 20 pixels below the face area.
- the clothes area is an area automatically calculated based on the face area, the area detected as the clothes area may extend out of the image as shown in the image 402. In this case, the clothes area is only the hatched area included in the image of the clothes area 402a '.
- the clothes area overlaps with the face area of another person as in the image 403
- the clothes area is an area not overlapping with the face area.
- two or more clothes areas overlap, it is considered that the person whose face area is detected below is taken in front of the camera, and the clothes of the person taken on the back side
- the area is an area that does not overlap with the clothes area of the person appearing in the front. This is based on the assumption that if the person in front is taller than the person in back, the person on the back is hidden by the person on the front and no face area is detected.
- the face area 403b is detected lower than the detected face areas 403a and 403b.
- the clothes area 403b 'corresponding to the face area 403b is preferentially detected.
- the clothes area corresponding to the face area 403a is a lattice area not overlapping with the clothes area 403b 'in the clothes area 403a' as the detection result.
- FIG. 5 shows an example of the face information detected by the face detection unit 103 from each image and the result of the clothes area detected by the clothes detection unit 104.
- An area surrounded by a solid line is a face area, and an area surrounded by a dotted line is a clothing area corresponding to the extracted face area.
- the clothing feature quantity extraction unit 105 extracts the image feature quantity of the clothing area detected by the clothing detection unit 104.
- the image feature amount indicates the feature of the distribution of pixel values related to a plurality of pixels in the image.
- the image feature amount of the clothes area may include, for example, the color content ratio in the clothes area, the amount of change in luminance between adjacent pixels, and the like.
- the clothes feature quantity extraction unit 105 registers the extracted image feature quantity of the clothes area in the person management information table 301.
- the clothes recognition unit 106 specifies the type of clothes based on the image feature amount of clothes in the person management information 301.
- the clothes recognition method is provided with a classifier that can identify the kind of clothes from the image feature amount by learning in advance using SVM (Support Vector Machine) method, and the kind of clothes corresponding to each face ID is Identify.
- the type of clothes recognized by the clothes recognition unit 106 is taken as first type information.
- the clothes recognition unit 106 registers the specified first type information in the person management information table 301. For example, in FIG. 3, the first type information of the person whose face ID is 2 or 3 is specified as gymnastic clothes. In addition, it may occur when the face ID is not specified as any clothes type, such as a person of one.
- the image event evaluation unit 107 evaluates an event of an image group based on the contents of the image management information table 201 and the contents of the person management information table 301. Detailed evaluation methods will be described later.
- an image group or an image is associated with an event.
- the storage unit 108 is an image management information table 201 including shooting date and time, an image group ID, and a face ID corresponding to each image ID, and person management information including an image feature amount of clothing and first type information corresponding to each face ID.
- a table 301, a table 801 of third type information and the number of images shown in FIGS. 8, 9 and 10 described later, a table 901 showing correspondence between clothes and events, and a table 1001 showing results of event evaluation of image groups are held. Do.
- the image evaluation apparatus 100 includes a processor and a memory (not shown), and the processor implements the respective functional units by executing a program stored in the memory. ⁇ Operation> Next, the operation of the present embodiment will be described using the flowchart shown in FIG.
- the image acquisition unit 101 acquires the images accumulated by the imaging device 120, and registers an image ID unique to each image in the image management information table 201 (step S601).
- the image group generation unit 102 generates an image group from the image group acquired by the image acquisition unit 101, and registers an image group ID unique to each group in the image management information table 201 (step S602).
- the face detection unit 103 detects a face area of a person from each image, and registers a face ID unique to each face in the image management information table 201 and the person management information table 301 (step S603).
- the clothes detection unit 104 detects clothes areas corresponding to the face areas from the face areas detected by the face detection unit 103 (step S604).
- the clothing feature quantity extraction unit 105 extracts the image feature quantity of the clothing area detected by the clothing detection unit 104, and registers it in the person management information table 301 (step S605).
- the clothes recognition unit 106 specifies the first type information of each person based on the image feature amount of clothes in the person management information table 301, and registers the specification result in the person management information table 301 (step S606).
- the image event evaluation unit 107 performs event evaluation of a plurality of image groups based on the contents of the image management information table 201 and the contents of the person management information table 301 created by executing the processing of steps S601 to S606 (steps S 607).
- FIG. 7 is a detailed flowchart of step S 607 showing an operation of the image event evaluation unit 107 performing event evaluation of one image group.
- the image event evaluation unit 107 selects, from the image group to be evaluated, an image group in which a person of a predetermined number or more is photographed (step S701).
- the image event evaluation unit 107 selects an image in which two or more persons are shown in order to perform event evaluation from clothes worn by a plurality of persons.
- the number of persons shown in the image can be known from the number of face IDs registered in the field of face ID in the image management information table 201.
- the image event evaluation unit 107 selects one image from the image group selected in step S701, and sets the type of clothing that characterizes the image (hereinafter, referred to as third type information) to the first reference in the image.
- the type of clothes satisfying the condition is identified (step S702).
- the first criterion is that the ratio of the number of the same first type information to the number of clothes included in the image exceeds 0.5. That is, if there is the same type of clothes worn by the majority of people appearing in the image, the image event evaluation unit 107 identifies the type of the clothes.
- step S702 A specific example of the operation in step S702 will be described using the image management information table 201 shown in FIG. 2 and the person management information table 301 shown in FIG.
- the image event evaluation unit 107 sets the image ID of the image selected in this step to 1.
- the image management information table 201 it is understood from the image management information table 201 that the number of persons included in the image is four with face IDs 1 to 4.
- the image event evaluation unit 107 specifies that the third type information for the image with an image ID of 1 is exercise clothes.
- the image event evaluation unit 107 counts the number of images characterized by the same third type information (step S703).
- the image event evaluation unit 107 creates the table 801 of the third type information and the number of images shown in FIG. 8 and increments the field of the number of images corresponding to the third type information specified in step S702. Count.
- the image event evaluation unit 107 determines whether the processing of all the images selected in step S701 is completed (step S704). In the case of YES, the process proceeds to step S705, and in the case of NO, the process returns to step S702.
- the image event evaluation unit 107 specifies the type of clothing that satisfies the second standard in the image group as the type of clothing that characterizes the image group (hereinafter, referred to as second type information) (step S705).
- the second criterion is that the ratio of the number of images characterized by the same third type information to the number of images selected in step S701 in the image group exceeds 0.5. That is, the image event evaluation unit 107 specifies the type of a large number of images among the images in which a plurality of persons in the image group is photographed, if there is clothing worn by the majority of persons photographed in the image.
- step S705 A specific example of the process of step S705 will be described using the third type information and the table 801 of the number of images shown in FIG.
- the number of images selected in step S701 in the image group is 50.
- the image event evaluation unit 107 performs event evaluation of the image group according to the second type information (step S706). Specifically, the image event evaluation unit 107 identifies an event associated with the second type information from the table 901 shown in FIG. 9 and stored in advance in the storage unit 108 and shown in FIG. Corresponds between the image group and the identified event. Also, an event tag of the specified event is attached to all the images belonging to the image group.
- the image evaluation apparatus 100 performs the above-described event evaluation on all image groups.
- FIG. 10 shows an example of the event evaluation result of the image group.
- the image evaluation device 100 outputs each image acquired by the image acquisition unit 101 to the display device 130 so that the evaluation result can be understood.
- the image evaluation device 100 combines each image and the event name of the attached event tag and outputs the result to the display device 130.
- the image evaluation apparatus 100 performs event evaluation of an image included in an image group based on the appearance frequency in the image group for each type of clothes with respect to an image group including two or more images.
- the image evaluation apparatus 100 according to the present embodiment Even if the image evaluation apparatus 100 according to the present embodiment incorrectly recognizes a plurality of clothes in a small number of images belonging to the above image group, if the clothes can be correctly recognized in most of the images belonging to the image group, I can evaluate it correctly. That is, the image evaluation apparatus 100 according to the present embodiment can perform evaluation more accurately than evaluation with one image.
- the event evaluation of the image is performed based on only the clothes information recognized by the clothes recognition unit 106.
- the similarity between clothes is calculated, and the event evaluation of the image is performed under the assumption that the clothes having high similarity are the same clothes type.
- the same parts as in the first embodiment will be assigned the same reference numerals as in the first embodiment, and the description thereof will be omitted.
- FIG. 11 is a functional block diagram of the image evaluation apparatus 1100 in the second embodiment.
- the image evaluation apparatus 1100 includes an image event evaluation unit 1102, a storage unit 1103, and an image information acquisition unit 1110 instead of the image event evaluation unit 107, the storage unit 108, and the image information acquisition unit 110 according to the first embodiment.
- the image information acquisition unit 1110 includes a similarity degree calculation unit 1101 in addition to the configuration of the image information acquisition unit 110.
- the similarity calculation unit 1101 calculates the similarity between clothes appearing in the same image based on the image feature amount of clothes managed by the person management information 301, and the clothes of the person shown in FIG. Is registered in the similarity information table 1201 between The similarity calculates cosine similarity between two vectors using image feature quantities of clothes as vectors.
- the image event evaluation unit 1102 performs event evaluation of the image group based on the contents of the image management information table 201, the contents of the person management information table 301, and the contents of the similarity information table 1201 between clothes of persons. Detailed evaluation methods will be described later.
- a storage unit 1103 includes an image management information table 201, a person management information table 301, a third type information / image number table 801, a table 901 showing correspondence between clothes and events, and a table 1001 showing image evaluation results of image groups.
- a similarity information table 1201 between clothes of persons in each image is held.
- the similarity calculation unit 1101 calculates the similarity between clothes appearing in the same image from the image feature amount of clothes in the person management information table 301 (step S1301).
- the image event evaluation unit 1102 performs event evaluation of a plurality of image groups based on the contents of the image management information table 201, the contents of the person management information table 301, and the contents of the similarity information table 1201 between clothes of persons ((1) Step S1302).
- FIG. 14 is a detailed flowchart of step S1301 showing an operation in which the image event evaluation unit 1102 evaluates an event of one image group.
- the image event evaluation unit 1102 selects, from the image group to be evaluated, an image group in which a person of a predetermined number or more is photographed (step S701).
- the image event evaluation unit 1102 selects one image from the plurality of images selected in step S701, and determines whether the combination of similar clothes among the clothes included in the image satisfies the third standard ( Step S1401). In the case of YES, the process proceeds to step S1403, and in the case of NO, the process proceeds to step S1402.
- the combination of similar clothes indicates a combination of two clothes whose similarity between clothes calculated by the similarity calculation unit 1101 exceeds 0.7.
- the third criterion is that, for example, the ratio of the number of combinations of similar clothes among the number of combinations of selecting two clothes from the clothes shown in the image exceeds 0.6.
- the evaluation unit 1102 determines that the combination of similar clothes does not satisfy the third standard.
- the image event evaluation unit 1102 determines whether the combination of similar clothes satisfies the fourth standard in the image selected in step S1401 (step S1402). In the case of Yes, the process proceeds to step S1403, and in the case of No, the process proceeds to step S1405.
- the fourth criterion is that, for example, assuming that the number of combinations of similar clothes is N, the average value of the similarities in the combinations of similar clothes exceeds the predetermined expression 0.9- (0.01 ⁇ N) To be.
- the image event evaluation unit 1102 identifies the type of clothing that satisfies the fifth standard as the type of clothing (third type information) that characterizes the image selected in the process of step S1401 (step S1403).
- the clothes recognition unit 106 identifies the kind of clothes meeting the fifth standard as a specific type It is assumed that the first type information exists.
- the same type is identified among the clothes included in the combination of all similar clothes in the image selected in the process of step S1401.
- the first type information having many numbers is referred to as third type information.
- step S1403 A specific example of the process of step S1403 will be described using the similarity information table 1201 between clothes of persons shown in FIG. 12 and the person management information table 301 shown in FIG.
- the combination of similar clothes is a face ID combination of (2, 3), (2, 4), (3, 4). It can be seen that the person wearing the clothes included in the combination of similar clothes has three face IDs of 2, 3, and 4. From the person management information table 301, the clothes of persons with face IDs of 2, 3 and 4 are identified as exercise clothes by the clothes recognition unit 106. Therefore, the image event evaluation unit 1102 specifies that the third type information is exercise clothes.
- the image event evaluation unit 1102 counts the number of images characterized by the same third type information (step S1404).
- the image event evaluation unit 1102 determines whether the processing of all the images selected in step S701 is completed (step S1405). In the case of YES, the process proceeds to step S1406, and in the case of NO, the process returns to step S1401.
- the image event evaluation unit 1102 specifies the type of clothing that satisfies the sixth standard in the image group as the type of clothing (second type information) that characterizes the image group (step S1406).
- the sixth criterion is that the ratio of the number of images characterized by the same third type information to the number of images selected in step S701 in the image group exceeds 0.5.
- the image event evaluation unit 1102 performs event evaluation of the image group according to the second type information (step S1407).
- the image evaluation device 1100 performs event evaluation on all image groups, and outputs each image acquired by the image acquisition unit 101 to the display device 130 so that an event of the evaluation result can be understood.
- the image evaluation device 1100 according to the second embodiment is an image group based on the appearance frequency for each type of clothes in the images included in the image group and the similarity between clothes for an image group consisting of two or more images. Evaluate the events contained in the image.
- the image evaluation apparatus 1100 according to the present embodiment may be able to specify exercise clothes that are not actually recognized as exercise clothes due to misrecognition as exercise clothes by using similarity of image feature amounts of clothes. There is. That is, the image evaluation device 1100 according to the present embodiment can perform more accurate evaluation than the image evaluation device 100 according to the first embodiment.
- Embodiment 3 In the first embodiment, the clothes of the person appearing in the image are estimated based on only the clothes information recognized by the clothes recognition unit 106, and the event evaluation of the image is performed.
- the same person shown in a plurality of images is identified by using clustering based on the feature amount of the face, and the same person wears the same clothes during the same event.
- FIG. 15 is a functional block diagram of the image evaluation device 1500 in the third embodiment.
- the image evaluation device 1500 is different from the image evaluation device 100 according to the first embodiment in the image event evaluation unit 107, the storage unit 108, and the image information acquisition unit 110 instead of the image event evaluation unit 1503, the storage unit 1504, the image information acquisition.
- a unit 1510 is provided.
- the image information acquisition unit 1510 includes a face feature amount extraction unit 1501 and a face clustering unit 1502 in addition to the configuration of the image information acquisition unit 110.
- a face feature amount extraction unit 1501 extracts an image feature amount of a face from the face area detected by the face detection unit 103.
- the extracted feature amounts of the face are managed in association with the face area.
- the face clustering unit 1502 performs clustering based on the image feature amounts of the face extracted by the face feature amount extraction unit 1501, and regards faces having similar image feature amounts of the face in the same image group as one cluster. Further, a unique cluster ID is assigned to each cluster, and is registered in a table 1601 indicating clusters for each face shown in FIG. 16 held by the storage unit 1504. It can be estimated that the persons of the faces classified into the same cluster are the same person.
- the image event evaluation unit 1503 evaluates an event of an image group based on the contents of the image management information table 201, the contents of the person management information table 301, and a table 1601 indicating clusters for each face. Detailed evaluation methods will be described later.
- the storage unit 1504 includes an image management information table 201, a person management information table 301, a third type information / image number table 801, a table 901 showing correspondence between clothes and events, and a table 1001 showing event evaluation results of image groups.
- a table 1601 showing clusters for each face, a table 1901 showing faces for each cluster shown in FIG. 19 described later, and a table 2001 showing faces belonging to the clusters shown in FIG. 20 and first type information are held.
- the face feature amount extraction unit 1501 extracts an image feature amount of each face from each face area detected by the face detection unit 103 (step S1701).
- the face clustering unit 1502 classifies faces with similar image feature amounts into one cluster based on the image feature amounts of the face extracted by the face feature amount extraction unit 1501 (step S1702).
- the face clustering unit 1502 assigns a unique cluster ID to each cluster, and registers the cluster ID in a table 1601 indicating clusters for each face.
- the image event evaluation unit 1503 evaluates the events of a plurality of image groups based on the contents of the image management information table 201, the contents of the person management information table 301, and the contents of the table 1601 indicating clusters for each face (step S1703). ).
- FIG. 18 is a detailed flowchart of step S1703 showing an operation of the image event evaluation unit 1503 evaluating an event of one image group.
- the image event evaluation unit 1503 selects, from the image group to be evaluated, an image group in which a person of a predetermined number or more is photographed (step S701).
- the image event evaluation unit 1503 selects one image to be determined from the plurality of images selected in step S701 (step S1801).
- the image event evaluation unit 1503 selects one face included in the image selected in step S1801 (step S1802).
- the image event evaluation unit 1503 extracts first type information specified by the clothing recognition unit 106 from each face belonging to the cluster, for the cluster to which the face selected in step S1802 belongs (step S1803).
- step S1803 will be specifically described using the table 1601 indicating clusters for each face.
- the face ID of the face selected in step S1802 is one.
- the cluster ID of the person whose face ID is 1 is 1.
- the image event evaluation unit 1503 extracts a record whose cluster ID is 1 from the table 1601 and creates a table 1901 indicating the face for each cluster shown in FIG.
- the face ID of the person whose cluster ID is 1 is 1, 13, 17 and 31 from the table 1901.
- First type information specified from faces with face IDs of 1, 13, 17 and 31 is extracted from the person management information table 301 of FIG.
- FIG. 20 shows an example of the extraction result.
- the image event evaluation unit 1503 identifies the type of clothing that satisfies the seventh standard as the type of clothing that characterizes the cluster to which the face selected in step S1802 belongs (hereinafter, referred to as fourth type information) (step S1804).
- the clothes worn by the person of the face selected in step S1802 are specified.
- the type of clothing that satisfies the seventh standard is the first type information that is identified as a specific type in the clothing recognition unit 106 among the first type information extracted in the process of step S1803.
- the first type information having a large number of being specified as the same type among the first type information extracted in the process of step S1803
- the fourth type information of the cluster to which the face selected in step S1802 belongs is used.
- the image event evaluation unit 1503 determines whether the processes of steps S1803 to S1804 have been completed for all the faces shown in the image selected in step S1801 (step S1805). If the determination is YES, the process advances to step S1806; if the determination is NO, the process returns to step S1802.
- the image event evaluation unit 1503 identifies the type of clothing that satisfies the eighth standard from the fourth type information for the face included in the image as clothing (third type information) that characterizes the image selected in step S1801 Step S1806).
- the eighth criterion is that, for example, the ratio of the number of pieces of fourth type information identical to the number of pieces of clothing identified in step S 1804 is 0 with respect to the number of clothes reflected in the image selected in step S 1801. .5 shall be exceeded.
- the image event evaluation unit 1503 counts the number of images characterized by the same third type information (step S1807).
- the image event evaluation unit 1503 determines whether the processing of all the images selected in step S701 is completed (step S1808). In the case of YES, the process proceeds to step S1809, and in the case of NO, the process returns to step S1801.
- the image event evaluation unit 1503 identifies the type of clothing that satisfies the ninth standard in the image group as the type of clothing (second type information) that characterizes the image group (step S1809).
- the ninth criterion is that the ratio of the number of images characterized by the same third type information to the number of images selected in step S701 in the image group exceeds 0.5.
- the image event evaluation unit 1503 performs event evaluation of the image group according to the second type information (step S1810).
- the image evaluation device 1500 performs event evaluation on all the image groups, and outputs each image acquired by the image acquisition unit 101 to the display device 130 so that the event of the evaluation result can be understood.
- the image evaluation device 1500 according to the third embodiment is not limited to the type of clothes in the image group included in the image group and the appearance frequency in the image group for each type of clothes and the face clustering result for the image group including two or more images. Based on the event evaluation of the image group.
- the image evaluation device 1500 wears a gym suit with another image even when a person wearing the gym clothes is not recognized as wearing a gym clothes in an image. If it can be recognized, the clothes worn by the person can be identified as the gym uniform even in the image which can not be recognized as wearing the gym uniform. That is, the image evaluation device 1500 of the present embodiment can perform more accurate evaluation than the image evaluation device 100 of the first embodiment.
- image event evaluation is performed based on the number of images in which a large number of persons wearing specific clothes are shown.
- FIG. 21 is a block diagram showing the configuration of the image evaluation device 2100 according to the third embodiment.
- the image evaluation device 2100 includes an image event evaluation unit 2101 and a storage unit 2102 instead of the image event evaluation unit 1503 and the storage unit 1504 of the configuration of the third embodiment.
- the image event evaluation unit 2101 performs event evaluation of the image group based on the contents of the image management information table 201, the contents of the person management information table 301, and the contents of the table 1601 indicating clusters for each face. Detailed evaluation methods will be described later.
- the storage unit 2102 includes an image management information table 201, a person management information table 301, a table 901 indicating correspondence between clothes and events, a table 1001 indicating an event evaluation result of image groups, a table 1601 indicating clusters for each face, and clusters A table 1901 indicating the face, a table 2001 indicating the face belonging to the cluster and the first type information, and a table 2401 of third type information and the number of clusters shown in FIG. 24 described later are held. ⁇ Operation> Next, the operation of this embodiment will be described using the flowchart shown in FIG. Here, since the processes of steps S601 to S606 and steps S1701 and S1702 are the same as in the third embodiment, the description will be omitted.
- the image event evaluation unit 2101 evaluates the events of a plurality of image groups based on the contents of the image management information table 201, the person management information table 301, and the table 1601 indicating clusters for each face (step S2201).
- FIG. 23 is a detailed flowchart of step S2201 showing an operation of the image event evaluation unit 2101 performing event evaluation of one image group.
- the image event evaluation unit 2101 selects one cluster in the image group to be evaluated (step S2301).
- the image event evaluation unit 2101 extracts the first type information specified by the clothes recognition unit 106 from each face belonging to the cluster selected in step S2301 (step S2302).
- the image event evaluation unit 2101 specifies the type of clothing that satisfies the tenth standard as clothing that characterizes the cluster selected in step S2301 (hereinafter, referred to as third type information) (step S2303).
- the type of clothing that satisfies the tenth standard is the first type information that is identified as a specific type in the clothing recognition unit 106 among the first type information extracted in step S2302.
- the first type information having a large number of being identified as the same type among the first type information extracted in step S2302 is selected in step S2301. It is assumed that the third type information of the selected cluster.
- the image event evaluation unit 2101 counts the number of clusters characterized by the same third type information (step S2304). That is, the image event evaluation unit 2101 counts, for each type of clothes, the number of persons wearing the same type of clothes in the image group.
- the image event evaluation unit 2101 creates a table 2401 of the number of clusters for each clothes shown in FIG. 24 and counts by incrementing the field of the number of clusters corresponding to the third type information specified in step S2303. .
- the image event evaluation unit 2101 determines whether the processing in steps S2302 to S2304 has been completed for all clusters in the image group (step S2305). In the case of Yes, the process proceeds to step S2306, and in the case of No, the process returns to step S2301.
- the image event evaluation unit 2101 specifies the type of clothing that meets the eleventh standard in the image group as the type of clothing (second type information) that characterizes the image group (step S2306).
- the eleventh criterion is that, for example, the ratio of the number of clusters characterized by the same third type information to the number of clusters in the image group exceeds 0.5. That is, if there is clothing worn by a majority of the persons appearing in the image group, the type is identified.
- step S2306 A specific example of the process of step S2306 will be described using the third type information and the table 2401 of the number of clusters shown in FIG.
- the number of clusters in an image group is ten.
- the image event evaluation unit 2101 performs event evaluation of the image group according to the second type information (step 2307).
- the image evaluation device 2100 performs event evaluation on all the image groups, and outputs each image acquired by the image acquisition unit 101 to the display device 130 so that the event of the evaluation result can be understood.
- the image evaluation device 2100 according to the fourth embodiment classifies the same person appearing in the image group into the same cluster for an image group consisting of two or more images, and uses the number of people wearing a specific clothes. Based on the event evaluation of the images included in the image group.
- image evaluation device 1500 of the third embodiment performs evaluation in units of images belonging to an image group, when a specific individual appears in a plurality of images, the person is evaluated according to the evaluation result as the number of images in which the person appears increases. It can have a major impact.
- image evaluation device 2100 of this embodiment the same person is evaluated in the same cluster, and evaluation is performed in cluster units, so that a specific individual does not greatly affect the evaluation result. It is possible to do event evaluation.
- groups are generated based on the shooting date and time acquired from the EXIF information, but the group generation method is not limited to this. For example, based on shooting points that can be acquired from metadata such as EXIF information, an image group may be generated as one group of images shot within a certain distance from a certain point.
- the clothes detection unit 104 detects a person whose face area is detected at the lower side as a person appearing in front of the camera
- the method of detecting the positional relationship of the person is this It is not limited to For example, a person whose face area is detected larger may be detected as a person who appears in front of the camera.
- the imaging device 120 can capture and store an image including parallax information such as a stereogram, the distance from the camera to the subject may be calculated based on the parallax information to detect the positional relationship of the person. Good.
- a classifier capable of specifying the kind of clothes according to the SVM method is provided in advance as a method for specifying the kind of clothes, but the present invention is not limited thereto.
- the type of clothes may be specified by matching the extracted image feature amount with the image feature amount serving as a template of each clothes.
- the image evaluation apparatus may further include an update information acquisition unit so as to acquire update information of these classifiers and templates via the network and update the classifiers and templates.
- This configuration enables the image evaluation device to change the type of clothes that can be identified as needed.
- the update information acquisition unit may acquire update information of the table 901 indicating the correspondence between clothes and events in accordance with changes in classifiers and templates, and may change an event that can be evaluated. According to this configuration, it is possible to evaluate the image group to an event corresponding to the newly specified type of clothes that can be identified.
- the similarity calculation unit 1101 calculates the cosine similarity between two vectors whose image feature amount of clothes is a vector as the similarity between clothes, but the similarity is limited to this. Absent. For example, the correlation coefficient of Pearson in the image feature amount between clothes, or the reciprocal of the Euclidean distance between two vectors of which the image feature amount of clothes is a vector may be calculated as the similarity.
- step S701 in the first and third embodiments an image in which only one person is photographed is excluded from the object of determination, but an image in which one person is photographed may be selected.
- Embodiment 2 Regarding the method of specifying the type of clothes that characterizes an image using the similarity calculated by the similarity calculation unit 1101, in Embodiment 2, the number of combinations of persons wearing similar clothes, or the similarity The type of clothes that characterizes the image is identified based on the average value of the similarities in the combination of the people wearing the clothes. However, the method of specifying the type of clothes that characterizes the image using the similarity is not limited to this.
- the image evaluation apparatus is characterized in that clothes having the same number of clothes that satisfy a certain standard as in the first embodiment as the clothes characterizing an image, based on the recognition result complemented by the above-described method It may be specified.
- the clothes that characterize the image group are identified by the number of images that are characterized by the same type of clothes.
- the method of identifying the clothing that characterizes the image group based on the type of clothing is not limited to this.
- each image may be weighted according to the number of persons shown in the image, and the weighted value may be added and evaluated for the number of images characterized by the same clothes type. According to the above-mentioned method, since the evaluation of the image in which many people appear is high, it is possible to identify the type of the dominant clothes in the clothes worn by many people.
- the type of clothes having a large number of appearances in one image is specified as the type of clothes that characterizes the image.
- the importance of the person appearing in the image may be calculated to identify the type of clothing that characterizes the image, taking into account the calculated importance.
- the importance of the person may be calculated based on, for example, the number of faces classified into each cluster, and the importance of the person appearing in many images may be calculated high, or the image may be calculated based on the appearance position or display size in the image. The importance of the person in the center or the person in the large area may be calculated high.
- the modification which specifies the kind of clothes which characterizes an image is described taking the case where the number of faces classified into each cluster is the importance of a person as an example.
- the face clustering unit 1502 manages the number of faces classified into each cluster and fourth type information (type of clothes that characterizes the cluster).
- fourth type information is specified by the image event evaluation unit 1503, but may be specified by the face clustering unit 1502.
- the number of faces of cluster 1 is “6”
- the type of clothes is “gymnastic clothes”
- the number of faces of cluster 2 is “2”
- the type of clothes is “no classification”
- the number of faces of cluster 3 Is 2 and the type of clothes is "Not classified”.
- the type of clothes that characterizes the image is as follows: Identify by the method shown in.
- the image event evaluation unit 1503 obtains cluster importance for each person (cluster) appearing in an image.
- the cluster importance is, for example, the number of faces of the cluster, the cluster importance of cluster 1 is 6, the cluster importance of cluster 2 is 2, and the cluster importance of cluster 3 is 2.
- the image event evaluation unit 1503 calculates image clothing importance for each type of clothing shown in the image.
- the image clothing importance is, for example, a normalized value of the cluster importance of clusters classified into the same clothing type.
- the image event evaluation unit 1503 identifies the type of clothing that meets the predetermined criteria (for example, the type of clothing whose image clothing importance exceeds 0.5) as the type of clothing that characterizes the image.
- the predetermined criteria for example, the type of clothing whose image clothing importance exceeds 0.5
- the image clothing importance level of the "gymnastic clothing” is 0.6 and exceeds 0.5
- the "gymnastic clothing” is identified as the type of clothing that characterizes the image.
- only the person with high importance may be used to specify the type of clothes that characterizes the image.
- a person of a cluster having a cluster importance of 4 or more is regarded as an important person.
- the number of faces in cluster 1 is “6”
- the type of clothing is “gymnastic clothes”
- the number of faces in cluster 2 is “2”
- the type of clothing is "no classification”
- the face of cluster 3 Assuming that the number is 2 and the type of clothes is "no classification", only important person in the image is cluster 1.
- the standard of clothing that characterizes an image is that the ratio of the type of a specific clothing among important persons appearing in the image exceeds 0.5.
- these variations can also be applied when evaluating the type of clothing that characterizes the image group. That is, weighting by cluster importance may be applied to evaluate the type of clothing characterizing the image group, or the type of clothing characterizing the image group may be specified using only the important persons appearing in the image group. Good.
- the importance of the person considered to be important to the photographer is calculated to be high, so that it is possible to perform event evaluation in accordance with the photographer's intention.
- the importance of the person shown in the image may be calculated based on the acquired information by acquiring information of an individual whose importance is to be increased from outside using SNS (Social Networking Service) or the like. For example, when face image data is acquired as information of an individual whose importance is to be increased from the outside, face feature amounts are extracted from the face image data, and matching is performed with face feature amounts of persons classified into each cluster. It is possible to calculate the importance of the person who wants to increase the importance.
- SNS Social Networking Service
- the image evaluation device acquires an image group from the imaging device 120 configured by a digital camera or the like, but the image acquisition destination is sufficient if there is a function of accumulating images, For example, an image group recorded in a recording medium such as a hard disk may be acquired.
- the image acquisition unit 101 acquires the image group stored in the imaging device 120 collectively, but the present invention is not limited to this.
- the corresponding image group may be acquired by specifying conditions such as shooting date and time from the image group stored in the imaging device 120.
- the image evaluation device performs event evaluation on each image group, and the display device 130 displays each image acquired by the image acquisition unit 101 as an evaluation result event.
- the usage of the evaluation result is not limited to this.
- a table indicating the event of the evaluation result of each image and the recording location (address) of the image file may be created and used as an index in the file system.
- the image evaluation device associates the image group and the event of the evaluation result on a one-to-one basis, but may associate the image group with a plurality of event candidates.
- a plurality of event candidates are associated with one clothes, and all the plurality of event candidates and the image group are associated with the second type information. You may associate.
- a plurality of pieces of second type information may be identified based on the number of occurrences of the third type information, for example, and may be associated with an event associated with each of the plurality of pieces of second type information. At this time, it may be displayed in a ranking format according to the number of appearances of the third type information.
- a table in which combinations of clothes and events are associated with each other such as a combination of "suit” and “dress” and “party” is prepared. Event evaluation of image groups may be performed.
- a control program comprising machine code or high-level language program code for causing the processor of the image evaluation apparatus and various circuits connected to the processor to execute the processes described in the first to fourth embodiments. It is also possible to record on a recording medium or distribute and distribute it via various communication paths and the like. Such recording media include an IC card, a hard disk, an optical disk, a flexible disk, a ROM, a flash memory and the like.
- the control program distributed and distributed is used by being stored in a memory or the like that can be read by a processor, and the processor executes the control program to realize each function as shown in each embodiment. Will be The processor may execute the control program directly, or may compile and execute or execute by the interpreter.
- Each functional component according to the first to fourth embodiments may be realized as an LSI (Large Scale Integration) which is an integrated circuit. These configurations may be individually made into one chip, or may be made into one chip so as to include part or all.
- LSI Large Scale Integration
- IC Integrated Circuit
- system LSI super LSI
- ultra LSI ultra LSI
- the method of circuit integration is not limited to LSI's, and integration may be performed using a dedicated circuit or a general purpose processor.
- a field programmable gate array that can be programmed after manufacturing the LSI, or a reconfigurable processor that can reconfigure connection and setting of circuit cells in the LSI may be used.
- the calculation of these functional blocks can be performed using, for example, a DSP (Digital Signal Processor) or a CPU (Central Processing Unit).
- these processing steps can also be processed by recording and executing as a program on a recording medium.
- the image evaluation apparatus includes: first specifying means for specifying, for each person shown in the image, first type information indicating the type of clothes worn by the person; Second specifying means for specifying second type information indicating a type of clothes characterizing the predetermined image group based on the appearance frequency for each type of first type information specified from a plurality of images belonging to the image group; And evaluation means for evaluating an event in which a plurality of images belonging to the predetermined image group are photographed based on the second type information.
- the image evaluation method includes a first identification step of identifying, for each person shown in the image, first type information indicating the type of clothes worn by the person; A second identification step of identifying second type information indicating a type of clothing that characterizes the predetermined image group based on the appearance frequency for each type of first type information identified from a plurality of images belonging to the image group; Evaluating an event in which a plurality of images belonging to the predetermined image group are photographed based on the second type information.
- a program is a program for causing a computer to execute an image evaluation process, and the image evaluation process is performed on clothes worn by the person for each person shown in the image.
- the predetermined image group is determined based on a first identification step of identifying first type information indicating the type of the image, and an appearance frequency for each type of first type information identified from a plurality of images belonging to the predetermined image group.
- the method further includes: a second identification step of identifying second type information indicating a type of clothing to be characterized; and an evaluation step of evaluating an event in which a plurality of images belonging to the predetermined image group are photographed based on the second type information.
- the integrated circuit includes: first specifying means for specifying, for each person shown in an image, first type information indicating the type of clothes worn by the person; Second specifying means for specifying second type information indicating the type of clothes characterizing the predetermined image group based on the appearance frequency for each type of first type information specified from the plurality of images belonging to the group; And evaluation means for evaluating an event in which a plurality of images belonging to the predetermined image group are photographed based on second type information.
- the image evaluation apparatus further specifies third type information indicating a type of clothes characterizing the image, based on the first type information specified from one image.
- a third identification unit is provided, and the second identification unit is configured to, based on the number of appearances of each type of the third type information identified for each of a plurality of images belonging to the predetermined group, a second for the predetermined image group
- the type information may be specified.
- the type of clothes that characterizes the plurality of images is identified by the number of images characterized by the same type of clothes, so the type of clothes appearing in many images among the plurality of images is specified. be able to.
- the third specifying unit may be configured such that the number of occurrences is a fixed ratio or more or a fixed number or more among the first type information specified from one image.
- One type of information may be specified as the third type of information.
- the image evaluation device of the above-mentioned embodiment (E) further extracts the image feature quantity of the clothes worn by the person shown in the image, and based on the image feature quantity, the clothes between the people shown in the image
- the third specifying means may specify the third type information based on the first type information specified from one image and the similarity degree.
- the third specifying unit is wearing similar clothes based on the similarity among the combinations of persons shown in one image.
- the combination of persons is selected, and when the number of combinations selected is equal to or greater than a certain percentage or more than a certain number, the first type information specified from each of the persons included in the selected combination is selected. Three types of information may be specified.
- the image evaluation device according to the above-mentioned embodiment (E) further extracts the feature amount of the face of the person shown in the image, and the same person shown in the plurality of images based on the similarity of the feature amounts of the face.
- fourth identification means for identifying fourth type information indicating the type of clothes that characterizes the cluster based on the first type information identified from the person belonging to each cluster.
- the third specifying unit may specify the third type information based on the fourth type information specified from a cluster of a person appearing in one image.
- clustering is performed according to the feature amount of the face to identify the same person shown in a plurality of images. Even if the type of clothes worn by a certain person is erroneously recognized in one image, there is a possibility that the type of clothes erroneously recognized from the recognition result of the clothes of the same person in another image can be corrected.
- the classification unit further calculates the importance of each of the clusters, and the third identification unit is identified from a cluster of a person appearing in one image.
- the third type information may be specified based on the fourth type information and the degree of importance.
- the image evaluation device of the above embodiment (A) further extracts feature amounts of the face of a person appearing in an image, and based on the similarity of the feature amounts of the face, the same person appearing in a plurality of images is extracted.
- the image processing apparatus further comprises: classification means for classifying into the same cluster; and third identification means for identifying third type information indicating the type of clothes characterizing the cluster based on first type information identified from a person belonging to the cluster.
- the second specification means may specify the second type information for the predetermined image group based on the number of appearances of the third type information in the predetermined image group.
- the type of clothes that characterizes a plurality of images is identified according to the number of clusters characterized by the same type of clothes, so that clothes worn by many persons among persons appearing in the plurality of images Can identify the type of In addition, it is possible to prevent individuals appearing in many images from significantly affecting evaluation results.
- the classification unit further calculates the importance of each of the clusters, and the second identification unit is configured to calculate the third type information in the predetermined image group.
- the second type information for the predetermined image group may be identified based on the number of occurrences of the image and the degree of importance.
- the event of the image group can be evaluated in consideration of the clothes of the important person.
- the first specification means specifies the first type information using clothes information for specifying the type of clothes; Furthermore, an update unit for updating the clothes information may be provided.
- the image evaluation apparatus further includes calculation means for detecting the area of the face of the person appearing in the image and calculating the area of clothes based on the area of the face;
- the specifying means may specify the first type information based on the image feature quantity extracted from the area of the clothes.
- the area of the face of the person shown in the image can be detected, the area of the clothes of the person can be detected.
- the calculation means specifies a person appearing in the front more than a plurality of detected persons based on the area of the face, and the overlapping clothes
- the area of the may be the area of the clothes of the person appearing in the front.
- the respective clothes areas can be determined based on the positional relationship of the person.
- the image evaluation apparatus can be applied to an apparatus for storing still images or moving pictures, digital cameras, photographing apparatuses such as mobile phones with cameras and movie cameras, and PCs (Personal Computers).
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Library & Information Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Processing Or Creating Images (AREA)
- Image Processing (AREA)
Abstract
Description
撮影者は、通常、1つのイベントに対し複数の画像を撮影する。そのような1つのイベントに対して撮影された複数の画像において、服装を正しく認識できた画像はそのイベントについて正しく評価できるが、服装を正しく認識できなかった画像はそのイベントについて正しく評価することが困難である。本発明は、服装を正しく認識できた画像の情報を用いることにより、服装を正しく認識できなかった画像についても、正しく評価することを可能とするものである。
<実施の形態1>
以下、本発明の一実施形態である画像評価装置100について図面を用いて説明する。
<構成>
図1は、実施の形態1における画像評価装置100の機能ブロック図である。図1に示すように、画像評価装置100は、撮影装置120および表示装置130と接続されている。
<動作>
次に、本実施形態の動作を図6に示すフローチャートを用いて説明する。
<まとめ>
本実施の形態の画像評価装置100は、2つ以上の画像からなる画像グループに対して、服装の種類ごとの画像グループにおける出現頻度に基づいて、画像グループに含まれる画像のイベント評価をする。
<実施の形態2>
実施の形態1では、服装認識部106で認識された服装情報のみに基づいて、画像のイベント評価を行っていた。実施の形態2では、実施の形態1に加えて、服装間の類似度を算出し、類似度の高い服装同士は同一の服装の種類であるという仮定のもとに、画像のイベント評価を行う方法を説明する。なお、構成とデータについて実施の形態1と同様の部分は同じ符号を付し説明を省略する。
<構成>
以下、本実施の形態の画像評価装置1100について説明する。図11は、実施の形態2における画像評価装置1100の機能ブロック図である。画像評価装置1100は、実施の形態1の画像イベント評価部107、記憶部108、画像情報取得部110の代わりに画像イベント評価部1102、記憶部1103、画像情報取得部1110を備える。画像情報取得部1110は、画像情報取得部110の構成に加えて、類似度算出部1101を備える。
<動作>
次に、本実施形態の動作を図13に示すフローチャートを用いて説明する。ここで、ステップS601~S606の処理は実施の形態1と同様であるので説明を省略する。
<まとめ>
実施の形態2の画像評価装置1100は、2つ以上の画像からなる画像グループに対して、画像グループに含まれる画像における服装の種類ごとの出現頻度および服装間の類似度に基づいて、画像グループに含まれる画像のイベント評価をする。
<実施の形態3>
実施の形態1では、服装認識部106で認識された服装情報のみに基づいて画像に写る人物の着ている服装を推定して、画像のイベント評価を行っている。実施の形態3では、実施の形態1に加えて、顔の特徴量に基づいたクラスタリングを利用して、複数の画像に写る同一人物を識別し、同一イベント中同一人物は同一の服装を着ているという仮定のもとに、画像のイベント評価を行う方法を説明する。なお、構成とデータについて実施の形態1と同様の部分は同様の符号を付し説明を省略する。
<構成>
以下、本発明の一実施形態である画像評価装置1500について説明する。図15は、実施の形態3における画像評価装置1500の機能ブロック図である。画像評価装置1500は、実施の形態1の画像評価装置100と比べて、画像イベント評価部107、記憶部108、画像情報取得部110の代わりに画像イベント評価部1503、記憶部1504、画像情報取得部1510を備える。画像情報取得部1510は画像情報取得部110の構成に加えて顔特徴量抽出部1501および顔クラスタリング部1502を備える。
<動作>
次に、本実施形態の動作を図17に示すフローチャートを用いて説明する。ここで、ステップS601~S606の処理は実施の形態1と同様の処理であるので説明を省略する。
<まとめ>
実施の形態3の画像評価装置1500は、2つ以上の画像からなる画像グループに対して、画像グループに含まれる画像における服装の種類と服装の種類ごとの画像グループにおける出現頻度および顔クラスタリング結果に基づいて、画像グループのイベント評価をする。
<実施の形態4>
実施の形態1~3では、特定の服装を着ている人物が多く写っている画像の数に基づいて、画像のイベント評価を行っている。実施の形態4では、実施の形態3における、顔クラスタリングを利用して、画像グループに登場する特定の種類の服装を着ている人物の数に基づいて、画像のイベント評価を行う方法を説明する。なお、構成とデータについて実施の形態1、3と同様の部分は同様の符号を付し説明を省略する。
<構成>
以下、本発明の一実施形態である画像評価装置2100について説明する。図21は、実施の形態3に係る画像評価装置2100の構成を示すブロック図である。画像評価装置2100は、実施の形態3の構成の画像イベント評価部1503、記憶部1504の代わりに画像イベント評価部2101、記憶部2102を備える。
<動作>
次に、本実施形態の動作を図22に示すフローチャートを用いて説明する。ここで、ステップS601~S606およびステップS1701、S1702の処理は実施の形態3と同様であるので説明を省略する。
<まとめ>
実施の形態4の画像評価装置2100は、2つ以上の画像からなる画像グループに対して、画像グループに登場する同一の人物を同一のクラスタに分類し、特定の服装を着ている人物数に基づいて、画像グループに含まれる画像のイベント評価をする。
<補足1>
上記実施形態について説明したが、本発明はこれに限られるものではない。以下、本発明の思想として含まれる各種変形例について説明する。
<補足2>
本発明の取り得る実施形態とその効果について説明する。
としてもよい。
更に、前記服装情報を更新する更新部を備えるとしてもよい。
101 画像取得部
102 画像グループ生成部
103 顔検出部
104 服装検出部
105 服装特徴量抽出部
106 服装認識部
107、1102、1503、2101 画像イベント評価部
108、1103、1504、2102 記憶部
110、1110、1510 画像情報取得部
120 撮影装置
130 表示装置
201 画像管理情報テーブル
301 人物管理情報テーブル
401、402、403 画像
402a、403a、403b 検出された顔領域
402a’、403a’、403b’ 検出された服装領域
801 第3種類情報と画像数を示すテーブル
901 服装とイベントの対応関係を示すテーブル
1001 画像グループごとのイベント評価結果を示すテーブル
1101 類似度算出部
1201 人物の服装間の類似度情報テーブル
1501 顔特徴量抽出部
1502 顔クラスタリング部
1601 顔ごとのクラスタを示すテーブル
1901 クラスタごとの顔を示すテーブル
2001 クラスタに属する顔と第1種類情報を示すテーブル
2401 第3種類情報とクラスタ数を示すテーブル
Claims (16)
- 画像に写る人物それぞれに対して、当該人物の着ている服装の種類を示す第1種類情報を特定する第1特定手段と、
所定の画像グループに属する複数の画像から特定される前記第1種類情報の種類ごとの出現頻度に基づいて、前記所定の画像グループを特徴付ける服装の種類を示す第2種類情報を特定する第2特定手段と、
前記第2種類情報に基づいて前記所定の画像グループに属する複数の画像が撮影されたイベントを評価する評価手段と
を備えることを特徴とする画像評価装置。 - 更に、1つの画像から特定される前記第1種類情報に基づいて、当該画像を特徴付ける服装の種類を示す第3種類情報を特定する第3特定手段を備え、
前記第2特定手段は、前記所定のグループに属する複数の画像ごとに特定される前記第3種類情報の種類ごとの出現数に基づいて、前記所定の画像グループに対する第2種類情報を特定する
ことを特徴とする請求項1記載の画像評価装置。 - 前記第3特定手段は、1つの画像から特定される前記第1種類情報のうち、出現数が一定割合以上または一定数以上である第1種類情報を、前記第3種類情報として特定する
ことを特徴とする請求項2に記載の画像評価装置。 - 更に、画像に写る人物の着ている服装の画像特徴量を抽出し、前記画像特徴量に基づいて前記画像に写る人物間の服装の類似度を算出する算出手段を備え、
前記第3特定手段は、1つの画像から特定される前記第1種類情報および前記類似度に基づいて、前記第3種類情報を特定する
ことを特徴とする請求項2に記載の画像評価装置。 - 前記第3特定手段は、1つの画像に写る人物の組合せのうち、前記類似度に基づいて類似する服装を着ていると判定される人物の組合せを選択し、選択された組合せの数が一定割合以上または一定数以上である場合に、前記選択された組合せに含まれる人物それぞれから特定される第1種類情報に基づいて、前記第3種類情報を特定する
ことを特徴とする請求項4に記載の画像評価装置。 - 前記第3特定手段は、1つの画像に写る人物の組合せのうち、前記類似度に基づいて類似する服装を着ていると判定される人物の組合せを選択し、選択された組合せにおける前記類似度の平均値が所定の式によって算出される値を超える場合に、前記選択された組合せに含まれる人物それぞれから特定される第1種類情報に基づいて、前記第3種類情報を特定する
ことを特徴とする請求項4に記載の画像評価装置。 - 更に、画像に写る人物の顔の特徴量を抽出し、前記顔の特徴量の類似性に基づいて、前記複数の画像に写る同一人物を同一のクラスタに分類する分類手段と、
各クラスタに属する人物から特定される第1種類情報に基づいて、当該クラスタを特徴付ける服装の種類を示す第4種類情報を特定する第4特定手段を備え、
前記第3特定手段は、1つの画像に写る人物のクラスタから特定される前記第4種類情報に基づいて、前記第3種類情報を特定する
ことを特徴とする請求項2に記載の画像評価装置。 - 前記分類手段は、更に、前記クラスタそれぞれの重要度を計算し、
前記第3特定手段は、1つの画像に写る人物のクラスタから特定される前記第4種類情報および前記重要度に基づいて、前記第3種類情報を特定する
ことを特徴とする請求項7に記載の画像評価装置。 - 更に、画像に写る人物の顔の特徴量を抽出し、前記顔の特徴量の類似性に基づいて、複数の画像に写る同一人物を同一のクラスタに分類する分類手段と、
前記クラスタに属する人物から特定される第1種類情報に基づいて、当該クラスタを特徴付ける服装の種類を示す第3種類情報を特定する第3特定手段を備え、
前記第2特定手段は、前記所定の画像グループにおける前記第3種類情報の出現数に基づいて、前記所定の画像グループに対する前記第2種類情報を特定する
ことを特徴とする請求項1に記載の画像評価装置。 - 前記分類手段は、更に、前記クラスタそれぞれの重要度を計算し、
前記第2特定手段は、前記所定の画像グループにおける前記第3種類情報の出現数および前記重要度に基づいて、前記所定の画像グループに対する前記第2種類情報を特定する
ことを特徴とする請求項9に記載の画像評価装置。 - 前記第1特定手段は、服装の種類を特定するための服装情報を用いて、前記第1種類情報を特定し、
更に、前記服装情報を更新する更新部を備える
ことを特徴とする請求項1に記載の画像評価装置。 - 更に、画像に写る人物の顔の領域を検出し、前記顔の領域に基づいて服装の領域を算出する算出手段を備え、
前記第1特定手段は、前記服装の領域から抽出される画像特徴量に基づいて前記第1種類情報を特定する
ことを特徴とする請求項1記載の画像評価装置。 - 前記算出手段は、前記顔の領域に基づいて、検出した複数の人物のうちより前面に写っている人物を特定し、重複する前記服装の領域を、より前面に写っている人物の服装の領域とする
ことを特徴とする請求項12記載の画像評価装置。 - 画像に写る人物それぞれに対して、当該人物の着ている服装の種類を示す第1種類情報を特定する第1特定ステップと、
所定の画像グループに属する複数の画像から特定される前記第1種類情報の種類ごとの出現頻度に基づいて、前記所定の画像グループを特徴付ける服装の種類を示す第2種類情報を特定する第2特定ステップと、
前記第2種類情報に基づいて前記所定の画像グループに属する複数の画像が撮影されたイベントを評価する評価ステップと
を含むことを特徴とする画像評価方法。 - コンピュータに画像評価処理を実行させるためのプログラムであって、
前記画像評価処理は、
画像に写る人物それぞれに対して、当該人物の着ている服装の種類を示す第1種類情報を特定する第1特定ステップと、
所定の画像グループに属する複数の画像から特定される第1種類情報の種類ごとの出現頻度に基づいて、前記所定の画像グループを特徴付ける服装の種類を示す第2種類情報を特定する第2特定ステップと、
前記第2種類情報に基づいて前記所定の画像グループに属する複数の画像が撮影されたイベントを評価する評価ステップと
を含むことを特徴とするプログラム。 - 画像に写る人物それぞれに対して、当該人物の着ている服装の種類を示す第1種類情報を特定する第1特定手段と、
所定の画像グループに属する複数の画像から特定される第1種類情報の種類ごとの出現頻度に基づいて、前記所定の画像グループを特徴付ける服装の種類を示す第2種類情報を特定する第2特定手段と、
前記第2種類情報に基づいて前記所定の画像グループに属する複数の画像が撮影されたイベントを評価する評価手段と
を備えることを特徴とする集積回路。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201280002838.3A CN103098088B (zh) | 2011-07-13 | 2012-07-05 | 图像评价装置、图像评价方法以及集成电路 |
US13/814,813 US9141856B2 (en) | 2011-07-13 | 2012-07-05 | Clothing image analysis apparatus, method, and integrated circuit for image event evaluation |
JP2013523814A JP5866360B2 (ja) | 2011-07-13 | 2012-07-05 | 画像評価装置、画像評価方法、プログラム、および集積回路 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2011155063 | 2011-07-13 | ||
JP2011-155063 | 2011-07-13 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2013008427A1 true WO2013008427A1 (ja) | 2013-01-17 |
Family
ID=47505736
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2012/004368 WO2013008427A1 (ja) | 2011-07-13 | 2012-07-05 | 画像評価装置、画像評価方法、プログラム、および集積回路 |
Country Status (4)
Country | Link |
---|---|
US (1) | US9141856B2 (ja) |
JP (1) | JP5866360B2 (ja) |
CN (1) | CN103098088B (ja) |
WO (1) | WO2013008427A1 (ja) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2017173900A (ja) * | 2016-03-18 | 2017-09-28 | 株式会社Nttドコモ | 情報処理装置 |
JP2021069067A (ja) * | 2019-10-28 | 2021-04-30 | 株式会社ソルトワークス | 画像処理方法及び画像処理プログラム |
Families Citing this family (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP5919665B2 (ja) * | 2011-07-19 | 2016-05-18 | 日本電気株式会社 | 情報処理装置、物体追跡方法および情報処理プログラム |
JP5967577B2 (ja) * | 2012-10-18 | 2016-08-10 | パナソニックIpマネジメント株式会社 | 共クラスタリング装置、共クラスタリング方法、プログラム及び集積回路 |
CN105469087B (zh) * | 2015-07-13 | 2017-04-19 | 百度在线网络技术(北京)有限公司 | 识别服饰图片的方法、服饰图片的标注方法及装置 |
US10523991B2 (en) * | 2015-08-31 | 2019-12-31 | Orcam Technologies Ltd. | Systems and methods for determining an emotional environment from facial expressions |
CN105260747B (zh) * | 2015-09-30 | 2019-07-23 | 广东工业大学 | 基于衣物共现信息与多任务学习的衣物识别方法 |
US20190057249A1 (en) * | 2016-02-26 | 2019-02-21 | Nec Corporation | Face recognition system, face matching apparatus, face recognition method, and storage medium |
US10136049B2 (en) * | 2017-01-09 | 2018-11-20 | International Business Machines Corporation | System, method and computer program product for contextual focus/zoom of event celebrities |
US10475195B2 (en) * | 2017-03-09 | 2019-11-12 | Intel Corporation | Automatic global non-rigid scan point registration |
JP7161318B2 (ja) * | 2018-06-20 | 2022-10-26 | 矢崎総業株式会社 | 乗車人数監視システム |
CN109801394B (zh) * | 2018-12-29 | 2021-07-30 | 南京天溯自动化控制系统有限公司 | 一种工作人员考勤方法及装置、电子设备和可读存储介质 |
US10726631B1 (en) * | 2019-08-03 | 2020-07-28 | VIRNECT inc. | Augmented reality system and method with frame region recording and reproduction technology based on object tracking |
US10949671B2 (en) * | 2019-08-03 | 2021-03-16 | VIRNECT inc. | Augmented reality system capable of manipulating an augmented reality object and an augmented reality method using the same |
US10719993B1 (en) * | 2019-08-03 | 2020-07-21 | VIRNECT inc. | Augmented reality system and method with space and object recognition |
CN112115756A (zh) * | 2020-03-22 | 2020-12-22 | 张冬梅 | 用于内容分析的区块链管理平台 |
CN112529871B (zh) * | 2020-12-11 | 2024-02-23 | 杭州海康威视系统技术有限公司 | 评价图像的方法、装置及计算机存储介质 |
CN113057529B (zh) * | 2021-02-22 | 2022-04-22 | 江苏柯林博特智能科技有限公司 | 一种基于楼梯清扫机器人的垃圾分类控制系统 |
CN113869435A (zh) * | 2021-09-30 | 2021-12-31 | 北京爱奇艺科技有限公司 | 图像处理、服饰识别方法、装置、设备及存储介质 |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009301119A (ja) * | 2008-06-10 | 2009-12-24 | Olympus Corp | 画像表示装置 |
JP2011517791A (ja) * | 2007-10-05 | 2011-06-16 | イーストマン コダック カンパニー | イベントマーカーとしての装飾 |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030174869A1 (en) * | 2002-03-12 | 2003-09-18 | Suarez Anthony P. | Image processing apparatus, image processing method, program and recording medium |
US7809722B2 (en) * | 2005-05-09 | 2010-10-05 | Like.Com | System and method for enabling search and retrieval from image files based on recognized information |
CN101079109B (zh) * | 2007-06-26 | 2011-11-30 | 北京中星微电子有限公司 | 基于制服特征的身份识别方法和系统 |
WO2011089884A1 (ja) | 2010-01-25 | 2011-07-28 | パナソニック株式会社 | 画像分類装置、方法、プログラム、プログラムを記録する記録媒体及び集積回路 |
-
2012
- 2012-07-05 WO PCT/JP2012/004368 patent/WO2013008427A1/ja active Application Filing
- 2012-07-05 US US13/814,813 patent/US9141856B2/en active Active
- 2012-07-05 JP JP2013523814A patent/JP5866360B2/ja active Active
- 2012-07-05 CN CN201280002838.3A patent/CN103098088B/zh active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011517791A (ja) * | 2007-10-05 | 2011-06-16 | イーストマン コダック カンパニー | イベントマーカーとしての装飾 |
JP2009301119A (ja) * | 2008-06-10 | 2009-12-24 | Olympus Corp | 画像表示装置 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2017173900A (ja) * | 2016-03-18 | 2017-09-28 | 株式会社Nttドコモ | 情報処理装置 |
JP2021069067A (ja) * | 2019-10-28 | 2021-04-30 | 株式会社ソルトワークス | 画像処理方法及び画像処理プログラム |
Also Published As
Publication number | Publication date |
---|---|
CN103098088A (zh) | 2013-05-08 |
US20130136313A1 (en) | 2013-05-30 |
US9141856B2 (en) | 2015-09-22 |
JP5866360B2 (ja) | 2016-02-17 |
JPWO2013008427A1 (ja) | 2015-02-23 |
CN103098088B (zh) | 2016-11-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2013008427A1 (ja) | 画像評価装置、画像評価方法、プログラム、および集積回路 | |
JP7317919B2 (ja) | 外観検索のシステムおよび方法 | |
CN109947975B (zh) | 图像检索装置、图像检索方法及其中使用的设定画面 | |
CN105118048B (zh) | 翻拍证件图片的识别方法及装置 | |
Zhao et al. | Learning mid-level filters for person re-identification | |
US9626551B2 (en) | Collation apparatus and method for the same, and image searching apparatus and method for the same | |
CN103747346B (zh) | 一种多媒体视频播放的控制方法及多媒体视频播放器 | |
JP5934653B2 (ja) | 画像分類装置、画像分類方法、プログラム、記録媒体、集積回路、モデル作成装置 | |
US9082013B2 (en) | Image recognition device, image recognition method, program, and integrated circuit | |
US9626585B2 (en) | Composition modeling for photo retrieval through geometric image segmentation | |
TWI712980B (zh) | 理賠資訊提取方法和裝置、電子設備 | |
JP2010532022A (ja) | 画像コレクションの合成人物モデル | |
CN107836109A (zh) | 电子设备自动聚焦于感兴趣区域的方法 | |
Lee et al. | Photographic composition classification and dominant geometric element detection for outdoor scenes | |
CN105740379A (zh) | 一种照片分类管理方法和装置 | |
Manyam et al. | Two faces are better than one: Face recognition in group photographs | |
JP2005148900A (ja) | 画像分類装置、画像分類方法、および、プログラム | |
CN109902550A (zh) | 行人属性的识别方法和装置 | |
JP7103229B2 (ja) | 不審度推定モデル生成装置 | |
CN106250396B (zh) | 一种图像标签自动生成系统及方法 | |
CN113920573B (zh) | 基于对抗学习的人脸变化解耦的亲属关系验证方法 | |
CN102314592B (zh) | 一种笑脸图像的识别方法和识别装置 | |
CN108052555A (zh) | 一种照片分类方法和系统 | |
CN104809435B (zh) | 一种基于视觉一致性约束的图像目标分类方法 | |
Pflug | Ear recognition: Biometric identification using 2-and 3-dimensional images of human ears |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 201280002838.3 Country of ref document: CN |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13814813 Country of ref document: US |
|
ENP | Entry into the national phase |
Ref document number: 2013523814 Country of ref document: JP Kind code of ref document: A |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 12811118 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: 12811118 Country of ref document: EP Kind code of ref document: A1 |