Image processing method, device, equipment and storage medium

Abstract

The present invention discloses an image processing method, device, equipment and storage medium. The method comprises: acquiring multiple frames of target images; including an image of a target object in each frame of the target image; using a feature determination model to determine the personality features corresponding to the target object in each frame of the target image; performing aggregation processing on the personality features corresponding to the target object in each frame of the target image to determine the target personality features corresponding to the target object in multiple frames of the target image. After determining the personality features of the target object in each frame of the target image, the present invention determines the target personality features by performing aggregation processing, and the target personality features are representative personality features shown by the target object in multiple frames of the image, which can fully show the original mental outlook of the target object.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image processing method, device, equipment and storage medium. Background Art

[0002] With the continuous advancement of science and technology, image processing technology continues to develop. Image processing technology can be applied to many application fields. For example, image processing technology can be used to identify and locate the target person in the video. Another example: image processing technology can be used to determine the status of the monitored object.

[0003] At present, most image processing technologies are implemented through image matching, that is, matching the image in the video data with a preset reference image. If the image in the video data successfully matches the reference image, it is determined that the target object (animal or person) corresponding to the reference image appears in the video data, and it is considered that the target object in the video data is in the state corresponding to the reference image. Among them, the reference image can be the morphological outline of the action that the target object may take, such as the curled up shape, the raised hand shape, etc. In other words, the current image processing technology can only analyze the morphological characteristics of the target object, and cannot analyze the individual characteristics of the target object, such as: it cannot reflect the temperament, image, personality, etc. of the target object.

[0004] However, in many application scenarios, it is necessary to determine the personality characteristics of the target object in order to determine the mental outlook of the target object. For example: in the process of artistic creation, casting is the biggest challenge. Many considerations are subjective and creative, but there are certain rules to follow, which are clear to experienced professional practitioners. Further, in the casting process, the actor's work is the most important basis for casting. Generally speaking, the total length of an actor's film and television works is often more than ten hours, or even hundreds of hours. Due to the lack of an image processing method for determining the personality characteristics of the target object, it is impossible to quickly analyze the video of the film and television works. The staff needs to spend a lot of time watching the actor's film and television works, so the staff needs to bear a huge workload when selecting actors who meet the characteristics of the role from a large number of candidate actors. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide an image processing method, apparatus, device and storage medium to solve the problem that image processing technology can only analyze the morphological characteristics of a target object but cannot analyze the individual characteristics of the target object.

[0006] In view of the above technical problems, the present invention is achieved through the following technical solutions:

[0007] In the first aspect of the implementation of the present invention, an image processing method is first provided, comprising: acquiring multiple frames of target images; wherein each frame of the target image includes an image of a target object; determining the individual features corresponding to the target object in each frame of the target image using a pre-trained feature determination model; performing aggregation processing on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image.

[0008] Wherein, the individual feature includes at least one feature value; each of the feature values ​​corresponds to a feature type; aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image, including: according to a preset number of reference feature types, aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image; wherein the number of feature values ​​in the target individual feature is less than or equal to the number of feature values ​​in the personality feature, and the feature type of each feature value in the target individual feature corresponds one-to-one to the preset number of reference feature types.

[0009] Among them, aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of target images, including: sequentially selecting image intervals in the multiple frames of target images using a preset sliding window; wherein each of the image intervals includes a portion of the target images in the multiple frames of target images; in each of the image intervals, aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image, and the aggregated identical individual features are used as candidate individual features; in multiple image intervals, aggregation processing is performed on the candidate individual features, and the target individual features are determined based on the aggregated identical candidate individual features.

[0010] Among them, before performing aggregation processing on the candidate personality features, it also includes: in each of the image intervals, determining the distribution density of the candidate personality features in the image interval; performing aggregation processing on the candidate personality features, including: in multiple image intervals, screening out candidate personality features whose distribution density is within a preset density range, and performing aggregation processing on the screened candidate personality features.

[0011] Among them, determining the distribution density of the candidate personality feature in the image interval includes: determining the edit distance between the candidate personality feature and each other personality feature in the image interval; determining the distribution density of the candidate personality feature based on the edit distance between the candidate personality feature and each other personality feature in the image interval, a preset distance threshold and the width of the sliding window.

[0012] Among them, determining the edit distance between the candidate personality feature and each of the remaining personality features in the image interval includes: for each of the personality features, by comparing the feature values ​​of the personality feature with those of the candidate personality features of each corresponding type, determining the distance value corresponding to each of the corresponding types; determining the sum of the distance values ​​corresponding to the corresponding types, and using the sum as the edit distance between the personality feature and the candidate personality feature.

[0013] Among them, the distribution density of the candidate personality feature is determined according to the edit distance between the candidate personality feature and each of the remaining personality features in the image interval, a preset distance threshold and the width of the sliding window, including: obtaining N personality features from the remaining personality features in the image interval except the candidate personality feature; wherein the edit distance between the N personality features and the candidate personality feature is less than the preset distance threshold; and the distribution density of the candidate personality feature is determined according to the frequency of occurrence of each of the N personality features in the image interval and the width of the sliding window.

[0014] Among them, there are multiple distance thresholds; determining the distribution density of the candidate personality traits includes: determining the distribution density of the candidate personality traits corresponding to each of the distance thresholds; and determining the distribution density of the candidate personality traits based on the distribution density of the candidate personality traits corresponding to each of the distance thresholds.

[0015] Among them, after determining the target personality feature corresponding to the target object in the multiple frames of target images, it also includes: determining the image interval corresponding to the target personality feature; in the image interval corresponding to the target personality feature, selecting video clips corresponding to a preset first number of image intervals as target personality video clips corresponding to the target object, and displaying the video display data corresponding to each of the target personality video clips in a preset first display device.

[0016] Among them, after determining the target personality characteristics corresponding to the target object in the multiple frames of target images, it also includes: determining the target image corresponding to the target personality characteristics; among the target images corresponding to the target personality characteristics, selecting a preset second number of target images as the target personality video images corresponding to the target object, and displaying each frame of the target personality video image in a preset first display device.

[0017] Among them, the target object is a target actor; after determining the target personality features corresponding to the target object in the multiple frames of target images, it also includes: obtaining keywords corresponding to the object to be matched; querying the feature words corresponding to the target personality features of the target object; determining the matching degree between the keywords corresponding to the object to be matched and the feature words corresponding to the target object.

[0018] Among them, the target object is a target actor, and the number of the target actors is multiple; the object to be matched is a role to be matched; determining the matching degree between the keywords corresponding to the object to be matched and the feature words corresponding to the target object, including: determining the matching degree between the keywords corresponding to the role to be matched and the feature words corresponding to each of the target actors, so as to perform an actor recommendation operation according to the matching degrees corresponding to the multiple target actors respectively.

[0019] Each of the feature types corresponds to at least two semantically related feature words, and the feature value of the feature type corresponds to one of the at least two feature words; or each of the feature types corresponds to one feature word, and the feature value of the feature type corresponds to the matching level between the target object and the feature word.

[0020] In the second aspect of the implementation of the present invention, there is also provided an image processing device, comprising: a receiving and acquiring module, configured to receive a plurality of frames of target images and acquire a plurality of frames of target images from the plurality of frames of target images; wherein each frame of the target image includes an image of a target object; a first determining module, configured to determine the individual features corresponding to the target object in each frame of the target image using a pre-trained feature determination model; and a second determining module, configured to perform aggregation processing on the individual features corresponding to the target object in each frame of the target image, and determine the target individual features corresponding to the target object in the plurality of frames of the target image.

[0021] In the third aspect of the implementation of the present invention, there is also provided an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; the memory is used to store computer programs; the processor is used to implement any of the method steps described above when executing the program stored in the memory.

[0022] In a fourth aspect of the implementation of the present invention, a computer-readable storage medium is further provided, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute any of the method steps described above.

[0023] In a fifth aspect of the implementation of the present invention, there is also provided a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any of the method steps described above.

[0024] The image processing method, device, equipment and storage medium provided by the embodiments of the present invention first use a feature determination model to determine the individual characteristics corresponding to the target object in each frame of the target image; and then determine the target individual characteristics corresponding to the target object in the video data through aggregation processing. Since the individual characteristics of the target object in one frame of the target image have limitations, the embodiments of the present invention determine the target individual characteristics corresponding to the target object in multiple frames of the target image by performing aggregation processing after determining the individual characteristics of the target object in each frame of the target image. The target individual characteristics are representative individual characteristics of the target object in multiple frames of images, which can fully show the original mental outlook of the target object. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0026] Figure 1 is a flowchart of an image processing method according to an embodiment of the present invention;

[0027] Figure 2 is a flow chart of steps for acquiring a target image according to an embodiment of the present invention.

[0028] Figure 3 is a schematic diagram of a model structure of a feature determination model according to an embodiment of the present invention;

[0029] Figure 4 is a flow chart of steps for determining target personality characteristics according to an embodiment of the present invention;

[0030] Figure 5 is a flow chart of steps for determining distribution density according to an embodiment of the present invention;

[0031] Figure 6 is a structural diagram of an image processing device according to an embodiment of the present invention;

[0032] Figure 7 is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the embodiment of the present invention will be described below in conjunction with the accompanying drawings in the embodiment of the present invention.

[0034] According to an embodiment of the present invention, an image processing method is provided. Figure 1 FIG. 1 is a flowchart of an image processing method according to an embodiment of the present invention.

[0035] Step S110, acquiring multiple frames of target images; wherein each frame of the target image includes an image of the target object.

[0036] The target image is an image in the video data.

[0037] The target object is the object of image processing. The types of target objects include but are not limited to: people and animals. For example, the target object can be a person or animal in real life, or a fictional character in a film or television work.

[0038] Specifically, video data is received and multiple frames of target images are acquired from the video data. The video data includes multiple consecutive frames of images.

[0039] Furthermore, the images in the video data may be sampled at a preset sampling interval; target object detection may be performed on the sampled images; the images in which the target object is detected may be used as the target images, or the images in which the target object is detected and the image quality meets the conditions may be used as the target images. After the target object is acquired, a target image sequence may be formed according to the playback position (playback time) of the target image in the video data. The image quality conditions will be introduced later and will not be described here in detail.

[0040] Furthermore, the sampling interval may be an empirical value or an experimental value. For example, if the video data is 25 frames per second, the sampling interval may be set based on sampling 8 frames per second.

[0041] Furthermore, a standard image of the target object can be pre-set, and the label of the standard image can be a unique code (qipuID) of the target object. In the video data, starting from the first frame image, the target object detection is performed on the first frame image, and after a sampling interval, the target object detection is performed on the next frame image until all the video data are detected. For the detected image, if the standard image of the target object is included in the image, the image is determined as the target image, or if the standard image is included in the image and the image meets the image quality condition, the image is determined as the target image.

[0042] Step S120, using the pre-trained feature determination model, determine the individual features corresponding to the target object in each frame of the target image.

[0043] The feature determination model is used to determine the personality characteristics of the target object in the target image.

[0044] Personality features are a collection of characteristic values ​​of the target object in a frame of image. Furthermore, personality features are used to represent the characteristic words of each dimension corresponding to the target object in the target image. The characteristic words are used to reflect the temperament characteristics, image characteristics, personality characteristics, etc. of the target object.

[0045] Specifically, the personality feature includes at least one feature value; each of the feature values ​​corresponds to a feature type. Each of the feature types corresponds to at least two semantically related feature words, and the feature value of the feature type corresponds to one of the at least two feature words; or, each of the feature types corresponds to a feature word, and the feature value of the feature type corresponds to the matching level between the target object and the feature word. Semantic relevance includes but is not limited to antonyms, that is, two semantically related feature words are antonyms of each other. Matching levels include but are not limited to primary matching, secondary matching, and tertiary matching with gradually increasing matching degrees. The feature types in the personality features can be determined according to requirements.

[0046] Personality traits can be represented by arrays or vectors. Each element position in the array corresponds to a feature type, and each dimension in the vector corresponds to a feature type. For example, 15 dimensions are set in personality traits. At least two semantically related feature words are set for each dimension. Feature words can be extracted from the character biography. The character biography is a concise text describing the character's characteristics, which contains key words, phrases, and other keywords that describe the character. The bag-of-words model can be used to extract key words, phrases, etc. from the character biography, and invalid descriptions that do not describe the character and express negative meanings can be removed using grammatical structure knowledge to obtain feature words. For example, the dimensions in personality traits include but are not limited to: dynamic-static, powerful-weak, immature-mature, and "dynamic", "static", "powerful", "weak", "immature", and "mature" are all feature words.

[0047] The feature determination model is a model obtained through pre-training. The purpose of training the feature determination model is that after the target image is input into the feature determination model, the feature determination model can determine and output the individual features of the target object in the target image. The training method of the feature determination model will be introduced later and will not be described here.

[0048] After training, the feature determination model can determine the operation value corresponding to each feature type, which indicates the position of the target object between the two antonyms corresponding to the feature type or the matching degree between the target object and a feature word corresponding to the feature type. The feature determination model can also discretize the operation value to obtain the polarity value corresponding to the operation value, which indicates the corresponding relationship between the target object and the two antonyms or the matching level between the target object and the feature word.

[0049] Polarity values ​​include -1, 0, and 1. When each feature type corresponds to two antonyms, -1 indicates that the target object belongs to the first feature word. 1 indicates that the target object belongs to the second feature word, and the first feature word and the second feature word are antonyms of each other. 0 indicates that the target object belongs to neither the first feature word nor the second feature word, but is between the first feature word and the second feature word. In this way, if 15 dimensions are set in the personality characteristics, the dimension values ​​of the 15 dimensions are connected together to form a 15-dimensional personality characteristic, which can theoretically represent 3 to the 15th power of personality characteristics. When each feature type corresponds to a feature word, -1 indicates a first-level match, 0 indicates a second-level match, and 1 indicates a third-level match.

[0050] Step S130, performing aggregation processing on the individual features corresponding to the target object in each frame of the target image, and determining the target individual features corresponding to the target object in the multiple frames of target data.

[0051] The target individual feature is a set of the characteristic values ​​of the target object in the video. The target individual feature can reflect the characteristic words of each characteristic type corresponding to the target object in the video data.

[0052] The aggregation process may include: performing a single aggregation process on the individual features corresponding to the target object in each frame of the target image, obtaining a set including the largest number of identical individual features, and using the individual features corresponding to the set as the target individual features; or performing multiple aggregation processes on the individual features corresponding to the target object in each frame of the target image, the multiple aggregation processes including: dividing the video data into multiple video sequence segments; for each video sequence segment, determining the candidate individual features corresponding to the target object in the video sequence segment by aggregating the individual features corresponding to the target object in each frame of the target image; determining the target individual features corresponding to the target object in the video data by aggregating the candidate individual features corresponding to the target object in each video segment. The candidate individual features may be the individual features aggregated in the largest number in the video sequence segment. The target individual features may be the candidate features aggregated in the largest number.

[0053] Of course, this embodiment is only used to describe the present invention, and the aggregation processing method is not limited to this. There may be other methods for determining the target personality characteristics. A more specific embodiment of other methods will be given later, and no further details will be given again.

[0054] In this embodiment, the characteristic determination model is first used to determine the individual characteristics corresponding to the target object in each frame of the target image; and then the target individual characteristics corresponding to the target object in the video data are determined through aggregation processing. Since the individual characteristics of the target object in one frame of the target image have limitations, the embodiment of the present invention determines the target individual characteristics corresponding to the target object in multiple frames of target images (multiple target images in the video data) by performing aggregation processing after determining the individual characteristics of the target object in each frame of the target image. The target individual characteristics are representative individual characteristics of the target object in multiple frames of target images, which can fully show the original mental outlook of the target object.

[0055] The following describes the steps for obtaining the target image.

[0056] In this embodiment, the target image in the video data is acquired according to the image detection technology and the image quality detection model, or the target image in the video data is acquired according to the image detection technology and the image quality condition.

[0057] Image detection technology can be used to identify whether there is an image of a target object in a video image.

[0058] The image quality detection model can be a deep learning network trained in advance, which is used to detect whether the input image meets the image quality condition. The image quality condition is used to measure whether the image quality can determine the individual characteristics so as to filter out candidate images with poor image quality.

[0059] The following describes the process of acquiring a target image in video data based on image detection technology and image quality conditions.

[0060] like Figure 2 FIG. 1 is a flowchart of steps for acquiring a target image according to an embodiment of the present invention.

[0061] Step S210, performing target object detection on each frame of image in the video data to obtain candidate images; wherein the candidate images include the image of the target object.

[0062] The image of the target object may be a head image of the target object or a full body image of the target object.

[0063] Step S220: determining the image quality index of the candidate image.

[0064] Image quality indicators include, but are not limited to: the area, angle, brightness and clarity of the image of the target object in the candidate image.

[0065] The area of ​​the target object's image refers to the area of ​​the target detection box corresponding to the target object. The target detection box is used to determine the position of the target object, and the image of the target object is within the target detection box.

[0066] The angle of the image of the target object refers to the angle of the target object. When the target object is a person, the angle of the target object is the side face or body angle of the target object. When the target object is an animal, the angle of the target object is the body angle of the target object. Further, the angle of the target object can be determined by an angle determination algorithm. For example, when the target object is a person, key points are identified in the image of the face, and the key points are mapped to the corresponding positions of the three-dimensional face model. At this time, the face angle of the three-dimensional face model is the angle of the face.

[0067] Step S230, obtaining a candidate image whose image quality index meets a preset image quality condition as a target image.

[0068] Image quality condition, used to filter candidate images with poor image quality.

[0069] The image quality conditions include, but are not limited to: a threshold value of the area ratio of the target object's image to the entire image, an angle range, a brightness range, and a clarity range. Each range end value in the image quality conditions may be an empirical value or an experimental value.

[0070] For example: the face image occupies an area greater than 1 / 100 of the entire image, the side angle of the face is less than 45 degrees, the brightness of the face image is greater than 50 lux (lx), and the clarity score of the face image is greater than 30 points.

[0071] In the candidate image, if the ratio of the area of ​​the target object's image to the entire image area is greater than the area ratio threshold, the target object's side face angle is within the angle range, the target object's image brightness is within the brightness range, and the target object's image clarity is within the clarity range, then the candidate image is determined to be the target image.

[0072] In this embodiment, after the image of the target object is detected in the image, the image quality index is evaluated for the image to ensure that when the personality feature is subsequently determined, the image quality of the target image will not affect the determination result of the personality feature.

[0073] After acquiring the target image from the video data, the individual characteristics of the target object in the target image can be determined. The following describes the steps of how to determine the individual characteristics of the target object in each frame of the target image.

[0074] The personality feature includes a feature value of at least one feature type, so that the feature determination model can be an image regression model that is independently trained for each feature type. Further, the type of the image regression model includes but is not limited to a deep convolutional neural network. The following is an example of a personality feature vector.

[0075] like Figure 3 FIG. 1 is a schematic diagram of a model structure of a feature determination model according to an embodiment of the present invention.

[0076] The model of the feature determination model includes M sub-models, M≥1, and the sub-model can be a classifier. The multiple sub-models share a bottom layer. Each sub-model corresponds to a dimension, and is used to determine and output the operation value of the dimension of the target object in the target image. The bottom layer is used to discretize the operation value into a polarity value. The bottom layer can be a Softmax layer.

[0077] A training sample set is pre-set, and the training sample set includes a plurality of sample images. Each sample image includes an image of a target object and is pre-labeled with a personality feature (multi-dimensional feature value). The personality feature is a real personality feature corresponding to the target object.

[0078] When training the feature determination model, the feature determination model can be trained in a supervised manner. The input of the feature determination model is a single sample image, and the output is a multidimensional feature value corresponding to the single sample image. Each dimension in the multidimensionality corresponds to a feature type. The multidimensional feature value is a predicted individual feature. The real multidimensional feature value of the input sample image is used to determine whether the feature determination model converges. After the feature determination model converges, the training of the feature determination model is stopped.

[0079] Furthermore, when training the feature determination model, the sub-model corresponding to each dimension can be trained independently. The sub-model of each dimension needs to be trained independently with multiple sample images (e.g., about 100,000 images) so that the sub-model can accurately determine the individual characteristics of the target object in each sample image.

[0080] Furthermore, each sub-model of the feature determination model outputs a calculation value of one dimension respectively, and the bottom layer of the feature determination model discretizes the calculation value of each dimension respectively, and discretizes the calculation value of each dimension into multiple polarity values.

[0081] For example: the initial feature determination model is an image regression model that includes 15 sub-models. The 15 sub-models are trained independently to obtain a feature determination model that can determine individual characteristics. Polarity values ​​include -1, 0, and 1. The range of the operation value output by the sub-model of each dimension is [-1,1]. The bottom layer of the feature determination model discretizes each operation value in a preset manner so that each operation value becomes one of the three polarity values ​​of -1, 0, and 1. Thresholds can be used to discretize the operation values, such as: the first threshold is 0.5, the second threshold is -0.5, when the operation value is greater than -0.5 and less than 0.5, the polarity value is 0, when the operation value is greater than or equal to 0.5, the polarity value is 1, and when the operation value is less than or equal to -0.5, the polarity value is -1.

[0082] In this embodiment, since the sub-model corresponding to each dimension needs to use more sample images for training, in order to achieve this sample size and ensure the annotation quality of the training data, the sample images can be selected according to the image quality conditions. Since the image quality conditions have been introduced above, they will not be repeated here. Furthermore, for the sub-model corresponding to each dimension, an image of the target object with significant personality characteristics (such as 1 or -1) in this dimension is obtained. For example: Use an image detection algorithm to detect images including target actors in film and television works, and filter out images that are too dark, have too large side face angles, are too small, or are too blurred.

[0083] When selecting sample images corresponding to each dimension, in order to further improve the quality of the sample images, obviously inappropriate video sequence segments in the video data can be filtered out. Inappropriate video sequence segments can be video sequence segments that do not match the personality characteristics, such as: segments of cross-dressing plots.

[0084] When selecting sample images corresponding to each dimension, it is necessary to control the number of classic characters to avoid too few classic characters or uneven distribution of sample images of different classic characters, which will lead to overfitting (overlearning) of personality characteristics of certain target objects (such as characters).

[0085] When selecting sample images corresponding to each dimension, sample images of different actors, genders, ages, etc. should be evenly distributed to avoid overfitting problems caused by unbalanced sample images.

[0086] After the feature determination model is trained, the feature determination model can be used to determine the individual features corresponding to the target object in each frame of the target image. According to the individual features corresponding to the target object in each frame of the target image, the target individual features of the target object in the video data are determined.

[0087] The feature types involved in the target personality feature can be determined according to user needs, and the number of feature types in the target personality feature is less than or equal to the number of feature types in the personality feature.

[0088] Specifically, according to a preset number of reference feature types, aggregation processing can be performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the video data; wherein the number of feature values ​​in the target individual features is less than or equal to the number of feature values ​​in the individual features, and the feature type of each feature value in the target individual features corresponds one-to-one to the preset number of reference feature types.

[0089] The reference feature type is the feature type to be present in the target personality feature. In other words, the feature type in the target personality feature is determined according to the reference feature type.

[0090] The reference feature type can be set according to the requirements. For example, if the personality feature includes the feature values ​​of 15 feature types, and the staff needs to determine the feature values ​​of multiple target feature types of the target person in the video, and the multiple target feature types are the first 7 feature types of the 15 feature types, then the first 7 feature types can be set as the reference feature type, so that the target personality feature can only include the feature values ​​of the 7 feature types.

[0091] The following describes the steps of how to determine the target personality characteristics through aggregation processing. Figure 4 FIG. 1 is a flowchart of steps for determining target personality characteristics according to an embodiment of the present invention.

[0092] Step S410, sequentially selecting image intervals in multiple frames of target images using a preset sliding window; wherein each of the image intervals includes a portion of the target images in the multiple frames of target images.

[0093] The width of the sliding window can be an empirical value or an experimental value.

[0094] Specifically, all acquired target images can be taken as an image sequence, and image intervals are sequentially selected in the image sequence using a sliding window, and each image interval is a video sequence segment.

[0095] The width of the sliding window may be a preset time length. Since the sampled target images are unevenly distributed in time, the number of target images in each sliding window may be different.

[0096] The width of the sliding window may be a preset number of images. Since the sampled target image is a segment of an image sequence, the sliding window may be used to divide the segment of the image sequence into a plurality of image intervals having the same number of images.

[0097] Step S420, in each of the image intervals, aggregation processing (first aggregation) is performed on the individual features of the target object corresponding to each frame of the target image, and the aggregated identical individual features are used as candidate individual features.

[0098] Among the individual features corresponding to the target images, the same individual features are aggregated and the aggregated same individual features are used as candidate individual features.

[0099] The number of candidate personality traits can be multiple. For example, after aggregation, the first personality trait has 3 targets, the second personality trait has 5 targets, and the third personality trait has 1 target. Then the first personality trait is a candidate personality trait, and the second personality trait is another candidate personality trait.

[0100] Step S430, performing aggregation processing (second aggregation) on candidate personality features in the plurality of image intervals, and determining the target personality feature based on the aggregated identical candidate personality features.

[0101] Determining the target personality feature according to the aggregated identical candidate personality features includes: determining feature values ​​corresponding to a preset number of reference feature types among the aggregated identical candidate personality features according to the preset number of reference feature types to form the target personality feature.

[0102] Furthermore, before performing aggregation processing on the candidate personality features, the distribution density of the candidate personality features in the image interval can also be determined in each of the image intervals; in multiple image intervals, candidate personality features with distribution density within a preset density range are screened out, and aggregation processing is performed on the screened candidate personality features, and the target personality feature is determined based on the aggregated identical candidate personality features.

[0103] Distribution density is used to reflect the frequency of candidate personality traits appearing per unit time. In image intervals with a relatively short time span, it reflects the concentration of personality traits.

[0104] Both end values ​​of the density range can be empirical or experimental values.

[0105] The selected candidate personality features are representative personality features in the image interval and are used to reflect the more significant personality features in the image interval.

[0106] The steps of how to determine the distribution density are further described below.

[0107] Each image interval includes multiple frames of target images, and each frame of the target image corresponds to a personality feature. The personality features in a single image interval are first aggregated to determine candidate personality features that can form a set.

[0108] like Figure 5 FIG. 1 is a flow chart of steps for determining distribution density according to an embodiment of the present invention. For each candidate personality feature in each image interval, the following steps are performed:

[0109] Step S510: determining the edit distance between the candidate individual feature and each of the remaining individual features in the image interval.

[0110] For an image interval, in each personality feature except the candidate personality feature, for each personality feature, by comparing the feature values ​​of the personality feature with those of the corresponding feature types of the candidate personality feature, determine the distance value corresponding to each corresponding feature type; determine the sum of the distance values ​​corresponding to the corresponding feature types, and use the sum as the edit distance between the personality feature and the candidate personality feature.

[0111] Specifically, when there are multiple candidate personality features in the image interval, the process of determining the edit distance for each candidate personality feature may adopt the following steps:

[0112] Step S1: for each personality feature except the candidate personality feature in the image interval, compare the characteristic value of the personality feature with the characteristic value of the corresponding characteristic type of the candidate personality feature, wherein the characteristic value is -1, 0 or 1.

[0113] Step S2: when the feature values ​​of the corresponding feature type are all -1 or all 1, determine that the distance value of the corresponding feature type is 0.

[0114] Step S3: when one of the feature values ​​of the corresponding feature type is 0, determine that the distance value of the corresponding feature type is 1.

[0115] Step S4, when one of the feature values ​​of the corresponding feature type is -1 and the other feature value is 1, determine that the distance value of the corresponding feature type is a preset value. The preset value is greater than 1. For example, the preset value is 1000.

[0116] Steps S2 to S4 are parallel steps. For example: The edit distance between candidate personality feature v1 and personality feature v2 = Σ1000 (when v1 i =-v2 i ), or 1(when v1 i *v2 i =0), or 0(when v1 i=v2 i ), where i represents the feature type.

[0117] Step S520, determining the distribution density of the candidate personality feature according to the edit distance between the candidate personality feature and each of the other personality features in the image interval, a preset distance threshold and the width of the sliding window.

[0118] In the image interval, N personality features are obtained from the remaining personality features except the candidate personality features; wherein the edit distance between the N personality features and the candidate personality features is less than a preset distance threshold; and the distribution density of the candidate personality features is determined according to the occurrence frequency of each of the N personality features in the image interval and the width of the sliding window. Specifically, the distribution density of the candidate personality features can be determined by the following formula:

[0119] D = count (V in image interval) ÷ ​​duration (T);

[0120] Wherein, D represents the distribution density of the candidate personality feature; V represents the candidate personality feature; count(V in image interval) represents the total occurrence frequency of personality features whose edit distance with V is less than the distance threshold in the edit distance between V and each other personality feature in the image interval; duration(T) represents the width of the sliding window.

[0121] Specifically, there are multiple distance thresholds, and each distance threshold corresponds to a level of distribution density. The distribution density (graded distribution density) of the candidate personality traits corresponding to each distance threshold is determined; and the distribution density (overall distribution density) of the candidate personality traits is determined according to the distribution density of the candidate personality traits corresponding to each distance threshold. The larger the distance threshold, the higher the level of distribution density, and the higher the level of distribution density, the rougher the degree of differentiation of personality traits.

[0122] Furthermore, the distribution density of the candidate personality traits is a linear combination of the graded distribution densities. For example, the graded distribution density is expressed as d(V,k), V is the candidate personality trait, K is the level corresponding to the graded distribution density, and the distribution density of the candidate personality traits d(V) is the weighted sum of the distribution densities of each level: d(V) = d(V,0) + 0.8d(V,1) + 0.5d(V,2) + 0.2d(V,3) + 0.1d(V,4). The weight corresponding to each level is an empirical value or an experimental value.

[0123] After determining the distribution density of each candidate personality feature in each image interval, all candidate personality features can be screened once, and candidate personality features with distribution density within a preset density range are retained. The retained candidate personality features are more representative, and an aggregation process is performed on all retained candidate personality features, the same candidate personality features are aggregated together to form a set, and the candidate personality features in each set are respectively used as a target personality feature. That is, the number of target personality features can be multiple.

[0124] In an embodiment of the present invention, each feature type in the personality feature corresponds to at least two semantically related feature words, and the feature value of the feature type corresponds to one of the at least two feature words. Since the target personality feature is determined based on the personality feature, each feature type in the target personality feature also corresponds to at least two semantically related feature words, and the feature value of the feature type corresponds to one of the at least two feature words. In this way, after determining the target personality feature, the feature words corresponding to the feature values ​​of each feature type in the target personality feature can be displayed in a preset display device. Furthermore, a feature type-feature value-feature word comparison table can be pre-set, so that the feature words corresponding to the feature values ​​of each feature type in the target personality feature can be determined by looking up the table.

[0125] Of course, when each feature type in the personality feature corresponds to a feature word, the feature value of the feature type corresponds to the matching level between the target object and the feature word. Since the target personality feature is determined based on the personality feature, each feature type in the target personality feature also corresponds to a feature word, and the feature value of the feature type also corresponds to the matching level between the target object and the feature word. In this way, after determining the target personality feature, the feature word and the matching level corresponding to the feature value of each feature type in the target personality feature can be displayed in a preset display device. Furthermore, a feature type-feature value-feature word comparison table and a feature type-feature value-matching level comparison table can be pre-set, so that the feature word and the matching level corresponding to the feature value of each feature type in the target personality feature can be determined by looking up the table.

[0126] In an embodiment of the present invention, in order to facilitate viewing of the significant manifestations of the target personality characteristics, after determining the target personality characteristics corresponding to the target object in the video data, the image interval corresponding to the target personality characteristics can be determined; in the image interval corresponding to the target personality characteristics, a preset first number of video segments corresponding to the image intervals are selected as the target personality video segments corresponding to the target object, and the video display data corresponding to each of the target personality video segments is displayed in a preset first display device. Each selected target personality video segment can present the personality characteristics of the target object in a targeted manner.

[0127] Furthermore, the video segment corresponding to the image interval refers to: in the video data, the video segment between the first frame target image and the last frame target image of the image interval.

[0128] Furthermore, when selecting an image interval, an image interval with the earliest playback time may be selected, and / or an image interval with the largest average value of the image quality index may be selected, and / or an image interval with the largest average value of the distribution density of the candidate personality feature may be selected, and the video segment corresponding to the selected image interval may be used as the target personality video segment. The target personality feature and the target personality video segment may be displayed on a preset first display device.

[0129] In an embodiment of the present invention, in order to facilitate viewing of representative images corresponding to target personality features, after determining the target personality features corresponding to the target object in the video data, a target image corresponding to the target personality features can be determined; among the target images corresponding to the target personality features, a preset second number of target images are selected as target personality video images corresponding to the target object, and each frame of the target personality video image is displayed on a preset second display device. The first display device and the second display device are the same or different display devices.

[0130] Furthermore, since the personality feature is determined by the feature determination model according to the target image, and the target personality feature is determined based on the personality feature, after determining the target personality feature, all target images corresponding to the target personality feature can be determined. Among all target images corresponding to the target personality feature, the target image with the earliest play time can be selected, and / or the target image with the largest image quality index can be selected, and / or the target image with the largest distribution density can be selected, and the selected target image can be used as the target personality video image corresponding to the target object. The target personality feature and the target personality video image can be displayed on a preset display device.

[0131] The image analysis method of the embodiment of the present invention has many application scenarios. For example, it can be applied in the scenario of intelligent casting, in the scenario of observing the personality (lively, calm) of pets, and in the scenario of recommending products according to the personality of customers.

[0132] Specifically, the keywords corresponding to the object to be matched may be obtained; the feature words corresponding to the target personality characteristics of the target object may be searched; and the matching degree between the keywords corresponding to the object to be matched and the feature words corresponding to the target object may be determined.

[0133] The matching degree can measure the degree of personality similarity between the target object and the object to be matched.

[0134] Furthermore, a preset feature word comparison table may be queried to determine the feature word corresponding to the target personality feature of the target object. The feature word comparison table is used to record the correspondence between the feature value and the feature word of each feature type. The feature word comparison table includes but is not limited to: a feature type-feature value-feature word comparison table and a feature type-feature word comparison table. When each feature type corresponds to at least two feature words, the feature type-feature value-feature word comparison table may be used. When each feature type corresponds to one feature word, the feature type-feature word comparison table may be used.

[0135] The following uses intelligent casting as an example for explanation.

[0136] The target object is a target actor. The object to be matched is a role to be matched.

[0137] The video data may be video work data of the target actor. In this case, the determined target personality characteristics are personality characteristics of the role played by the target actor.

[0138] After determining the target personality features corresponding to the target object in the multiple frames of target images, the keywords corresponding to the role to be matched can be obtained; the matching degree between the keywords corresponding to the role to be matched and the target personality features corresponding to the target actor can be determined. The similarity between the role to be matched and the target actor can be determined based on the matching degree.

[0139] The embodiment of the present invention can also be used to recommend actors. Specifically, the target object is a target actor, and the number of the target actors is multiple. The object to be matched is a role to be matched. After determining the target personality features corresponding to the target object in the multiple frames of target images, the keywords corresponding to the role to be matched can be obtained; the matching degree between the keywords corresponding to the role to be matched and the target personality features corresponding to each of the target actors is determined, so as to perform an actor recommendation operation according to the matching degrees corresponding to the multiple target actors.

[0140] Furthermore, the actor recommendation operation includes: sorting the matching degrees corresponding to the plurality of target actors in descending order of matching degree, and displaying the sorting results on a preset display device for reference by casting personnel.

[0141] Furthermore, after determining the matching degree between the target actor and the role to be matched, the target actors whose matching degree is greater than a preset matching degree threshold can be screened out, and the feature words corresponding to the feature values ​​of each feature type in the target personality features corresponding to the target actor can be displayed in a preset display device, thereby realizing casting recommendations.

[0142] Of course, the target personality video clip and the target personality video image corresponding to the target actor may also be displayed together on the display device.

[0143] Furthermore, before determining the target personality characteristics corresponding to the target object in the video data, the role to be matched and the keywords corresponding to the role to be matched are determined; according to the keywords corresponding to the role to be matched, a reference feature type is set so that the reference feature type corresponding to the keyword is included in the determined target personality characteristics, so that the feature values ​​in the keywords corresponding to the role to be matched and the target personality characteristics are comparable.

[0144] In this embodiment, during the intelligent casting process, this embodiment starts from the basic principles of casting, uses algorithms to interpret the actors' works, and derives the generalized characteristics of the target actors (target personality characteristics) therefrom. Representative clips (target personality video clips) and moments (target personality video images) can also be selected, and the subjective character temperament is quantified. The personality of the character in the video data is understood through the algorithm, which saves casting time and improves the efficiency of casting.

[0145] By applying the embodiments of the present invention to the intelligent casting scenario, an integrated artist information release and screening platform can be built in terms of online video distribution, film industry, contracted artists, self-media, publicity and promotion teams, etc., to improve the objectivity of decision-making in the casting process and improve the efficiency and quality of casting. At the same time, it can also provide a direct batch channel for personal information for top stars, a large number of second- and third-tier artists, and newcomers in the performing arts, further narrowing the imbalance between artist status and resources, improving the efficiency of filming and matching, completing the effective utilization and allocation of resources, and maximizing value.

[0146] The embodiment of the present invention also provides an image processing device. Figure 6 FIG. 4 is a structural diagram of an image processing device according to an embodiment of the present invention.

[0147] The image processing device includes: a receiving and acquiring module 610 , a first determining module 620 and a second determining module 630 .

[0148] The receiving and acquiring module 610 is used to acquire multiple frames of target images, wherein each frame of the target image includes an image of a target object.

[0149] The first determination module 620 is used to determine the individual features corresponding to the target object in each frame of the target image by using a pre-trained feature determination model.

[0150] The second determination module 630 is used to perform aggregation processing on the individual features corresponding to the target object in each frame of the target image, and determine the target individual features corresponding to the target object in the multiple frames of the target image.

[0151] The functions of the device described in the embodiment of the present invention have been described in the above method embodiment. Therefore, for details not provided in the description of the embodiment of the present invention, reference may be made to the relevant description in the above embodiment and will not be repeated here.

[0152] The embodiment of the present invention further provides an electronic device, such as Figure 7 As shown, it includes a processor 710 , a communication interface 720 , a memory 730 and a communication bus 740 , wherein the processor 710 , the communication interface 720 , and the memory 730 communicate with each other via the communication bus 740 .

[0153] The memory 730 is used to store computer programs.

[0154] The processor 710 is used to execute the program stored in the memory 730 to implement the following steps:

[0155] Acquire multiple frames of target images; wherein each frame of the target image includes an image of a target object; determine the individual features corresponding to the target object in each frame of the target image using a pre-trained feature determination model; perform aggregation processing on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image.

[0156] Wherein, the individual feature includes at least one feature value; each of the feature values ​​corresponds to a feature type; aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image, including: according to a preset number of reference feature types, aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image; wherein the number of feature values ​​in the target individual feature is less than or equal to the number of feature values ​​in the personality feature, and the feature type of each feature value in the target individual feature corresponds one-to-one to the preset number of reference feature types.

[0157] Among them, aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of target images, including: sequentially selecting image intervals in the multiple frames of target images using a preset sliding window; wherein each of the image intervals includes a portion of the target images in the multiple frames of target images; in each of the image intervals, aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image, and the aggregated identical individual features are used as candidate individual features; in multiple image intervals, aggregation processing is performed on the candidate individual features, and the target individual features are determined based on the aggregated identical candidate individual features.

[0158] Among them, before performing aggregation processing on the candidate personality features, it also includes: in each of the image intervals, determining the distribution density of the candidate personality features in the image interval; performing aggregation processing on the candidate personality features, including: in multiple image intervals, screening out candidate personality features whose distribution density is within a preset density range, and performing aggregation processing on the screened candidate personality features.

[0159] Among them, determining the distribution density of the candidate personality feature in the image interval includes: determining the edit distance between the candidate personality feature and each other personality feature in the image interval; determining the distribution density of the candidate personality feature based on the edit distance between the candidate personality feature and each other personality feature in the image interval, a preset distance threshold and the width of the sliding window.

[0160] Among them, determining the edit distance between the candidate personality feature and each of the remaining personality features in the image interval includes: for each of the personality features, by comparing the feature values ​​of the personality feature with those of the candidate personality features of each corresponding type, determining the distance value corresponding to each of the corresponding types; determining the sum of the distance values ​​corresponding to the corresponding types, and using the sum as the edit distance between the personality feature and the candidate personality feature.

[0161] Among them, the distribution density of the candidate personality feature is determined according to the edit distance between the candidate personality feature and each of the remaining personality features in the image interval, a preset distance threshold and the width of the sliding window, including: obtaining N personality features from the remaining personality features in the image interval except the candidate personality feature; wherein the edit distance between the N personality features and the candidate personality feature is less than the preset distance threshold; and the distribution density of the candidate personality feature is determined according to the frequency of occurrence of each of the N personality features in the image interval and the width of the sliding window.

[0162] Among them, there are multiple distance thresholds; determining the distribution density of the candidate personality traits includes: determining the distribution density of the candidate personality traits corresponding to each of the distance thresholds; and determining the distribution density of the candidate personality traits based on the distribution density of the candidate personality traits corresponding to each of the distance thresholds.

[0163] Among them, after determining the target personality feature corresponding to the target object in the multiple frames of target images, it also includes: determining the image interval corresponding to the target personality feature; in the image interval corresponding to the target personality feature, selecting video clips corresponding to a preset first number of image intervals as target personality video clips corresponding to the target object, and displaying the video display data corresponding to each of the target personality video clips in a preset first display device.

[0164] Among them, after determining the target personality characteristics corresponding to the target object in the multiple frames of target images, it also includes: determining the target image corresponding to the target personality characteristics; among the target images corresponding to the target personality characteristics, selecting a preset second number of target images as the target personality video images corresponding to the target object, and displaying each frame of the target personality video image in a preset second display device.

[0165] Among them, after determining the target personality features corresponding to the target object in the multiple frames of target images, it also includes: obtaining keywords corresponding to the object to be matched; querying the feature words corresponding to the target personality features of the target object; and determining the degree of matching between the keywords corresponding to the object to be matched and the feature words corresponding to the target object.

[0166] Among them, the target object is a target actor, and the number of the target actors is multiple; the object to be matched is a role to be matched; determining the matching degree between the keywords corresponding to the object to be matched and the feature words corresponding to the target object, including: determining the matching degree between the keywords corresponding to the role to be matched and the feature words corresponding to each of the target actors, so as to perform the actor recommendation operation according to the matching degrees corresponding to the multiple target actors.

[0167] Each of the feature types corresponds to at least two semantically related feature words, and the feature value of the feature type corresponds to one of the at least two feature words; or each of the feature types corresponds to one feature word, and the feature value of the feature type corresponds to the matching level between the target object and the feature word.

[0168] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0169] The communication interface is used for communication between the above terminal and other devices.

[0170] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0171] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0172] In another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions. When the instructions are executed on a computer, the computer executes the image processing method described in any one of the above embodiments.

[0173] In another embodiment of the present invention, a computer program product including instructions is provided. When the computer program product is run on a computer, the computer executes the image processing method described in any one of the above embodiments.

[0174] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0175] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0176] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0177] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. An image processing method, characterized in that: include: Acquire multiple frames of target images; wherein each frame of the target image includes a target object; Using the pre-trained feature determination model, determine the personality feature corresponding to the target object in each frame of the target image; wherein the personality feature includes at least one feature value; each of the feature values ​​corresponds to a feature type; each of the feature types corresponds to at least two semantically related feature words, and the feature value of the feature type corresponds to one of the at least two feature words; or each of the feature types corresponds to a feature word, and the feature value of the feature type corresponds to the matching level between the target object and the feature word; the semantic relevance includes: antonyms; Aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image.

2. The method according to claim 1, characterized in that Performing aggregation processing on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image includes: According to a preset number of reference feature types, aggregation processing is performed on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image; wherein the number of feature values ​​in the target individual features is less than or equal to the number of feature values ​​in the individual features, and the feature type of each feature value in the target individual features corresponds one-to-one to the preset number of reference feature types.

3. The method according to claim 1 or 2, characterized in that: Performing aggregation processing on the individual features corresponding to the target object in each frame of the target image to determine the target individual features corresponding to the target object in the multiple frames of the target image includes: Sequentially selecting image intervals in the multiple frames of target images using a preset sliding window; wherein each of the image intervals includes a portion of the target images in the multiple frames of target images; In each of the image intervals, performing aggregation processing on the individual features of the target object corresponding to each of the target images in each frame, and taking the aggregated identical individual features as candidate individual features; In the plurality of image intervals, aggregation processing is performed on candidate personality features, and the target personality feature is determined based on the aggregated identical candidate personality features.

4. The method according to claim 3, characterized in that: Before performing aggregation processing on the candidate personality features, the method further includes: In each of the image intervals, determining the distribution density of the candidate personality features in the image interval; Performing aggregation processing on the candidate personality features includes: In the plurality of image intervals, candidate individual features whose distribution density is within a preset density range are screened out, and aggregation processing is performed on the screened candidate individual features.

5. The method according to claim 4, characterized in that Determining the distribution density of the candidate personality feature in the image interval includes: Determine the edit distance between the candidate personality feature and each of the remaining personality features in the image interval; The distribution density of the candidate personality feature is determined according to the edit distance between the candidate personality feature and each of the other personality features in the image interval, a preset distance threshold and the width of the sliding window.

6. The method according to claim 5, characterized in that Determining the edit distance between the candidate personality feature and each of the remaining personality features in the image interval includes: For each of the personality characteristics, by comparing the characteristic values ​​of the personality characteristics with the characteristic values ​​of the corresponding characteristic types of the candidate personality characteristics, a distance value corresponding to each of the corresponding characteristic types is determined; The sum of the distance values ​​corresponding to the corresponding feature types is determined, and the sum is used as the edit distance between the personality feature and the candidate personality feature.

7. The method according to claim 5, characterized in that Determining the distribution density of the candidate personality feature according to the edit distance between the candidate personality feature and each of the remaining personality features in the image interval, a preset distance threshold, and the width of the sliding window, including: Obtaining N personality features from the remaining personality features in the image interval except the candidate personality feature; wherein the edit distances between the N personality features and the candidate personality feature are less than a preset distance threshold; The distribution density of the candidate personality features is determined according to the occurrence frequency of each of the N personality features in the image interval and the width of the sliding window.

8. The method according to any one of claims 5 to 7, characterized in that: The number of the distance thresholds is multiple; Determining the distribution density of the candidate personality traits includes: Determine the distribution density of the candidate personality features corresponding to each of the distance thresholds; The distribution density of the candidate personality feature is determined according to the distribution density of the candidate personality feature corresponding to each of the distance thresholds.

9. The method according to any one of claims 4 to 7, characterized in that: After determining the target individual features corresponding to the target object in the multiple frames of target images, the method further includes: Determine the image interval corresponding to the target personality feature; In the image interval corresponding to the target personality feature, video clips corresponding to a preset first number of image intervals are selected as target personality video clips corresponding to the target object, and video display data corresponding to each target personality video clip is displayed in a preset first display device.

10. The method according to any one of claims 1-2 and 4-7, characterized in that: After determining the target individual features corresponding to the target object in the multiple frames of target images, the method further includes: Determine a target image corresponding to the target personality feature; Among the target images corresponding to the target personality feature, a preset second number of target images are selected as target personality video images corresponding to the target object, and each frame of the target personality video image is displayed on a preset second display device.

11. The method according to any one of claims 1-2 and 4-7, characterized in that: After determining the target individual features corresponding to the target object in the multiple frames of target images, the method further includes: Get the keywords corresponding to the object to be matched; Querying the characteristic words corresponding to the target personality characteristics of the target object; Determine the degree of match between the keyword corresponding to the to-be-matched object and the feature word corresponding to the target object.

12. The method according to claim 11, characterized in that The target object is a target actor, and the number of the target actors is multiple; The object to be matched is a role to be matched; Determining the degree of match between the keyword corresponding to the to-be-matched object and the feature word corresponding to the target object includes: The matching degree between the keyword corresponding to the role to be matched and the feature word corresponding to each target actor is determined, so as to perform an actor recommendation operation according to the matching degrees corresponding to the multiple target actors.

13. An image processing device, characterized in that: include: A receiving and acquiring module, used for acquiring multiple frames of target images; wherein each frame of the target image includes an image of a target object; The first determination module is used to determine the individual features corresponding to the target object in each frame of the target image by using a pre-trained feature determination model; wherein the individual features include at least one feature value; each of the feature values ​​corresponds to a feature type; each of the feature types corresponds to at least two semantically related feature words, and the feature value of the feature type corresponds to one of the at least two feature words; or each of the feature types corresponds to a feature word, and the feature value of the feature type corresponds to a matching level between the target object and the feature word; the semantic relevance includes: antonyms; The second determination module is used to perform aggregation processing on the individual features corresponding to the target object in each frame of the target image, and determine the target individual features corresponding to the target object in the multiple frames of the target image.

14. An electronic device, characterized in that: include: A processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; A processor, for implementing the method steps described in any one of claims 1-12 when executing a program stored in a memory.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method steps described in any one of claims 1 to 12 are implemented.

Images