CN110909654A

CN110909654A - Training image generation method and device, electronic equipment and storage medium

Info

Publication number: CN110909654A
Application number: CN201911128049.4A
Authority: CN
Inventors: 王露; 朱烽; 赵瑞
Original assignee: Shenzhen Sensetime Technology Co Ltd
Current assignee: Shenzhen Sensetime Technology Co Ltd
Priority date: 2019-11-18
Filing date: 2019-11-18
Publication date: 2020-03-24

Abstract

The disclosure relates to a method and a device for generating a training image, an electronic device and a storage medium, wherein the method comprises the following steps: acquiring an original training image; randomly selecting a sub-set of the added objects in the added object set; obtaining a target adding object according to the adding object subset; and according to the target addition object, carrying out image processing on the original training image to obtain a training image. Through the process, the existing original training image can be utilized, the object is added based on the randomly selected target, so that a large number of training images added with the target addition object can be obtained relatively simply, and the target addition object is obtained based on the randomly selected addition object subset, so that the reliability, the reality and the diversity of the training images can be further improved along with the difference of the types and the number of the addition object subsets.

Description

Training image generation method and device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of image processing technologies, and in particular, to a method and an apparatus for generating a training image, an electronic device, and a storage medium.

Background

The existing neural network algorithm has low recognition accuracy when the images with shielding are recognized, and the recognition performance of the neural network can be effectively improved by training the neural network model through the image data with shielding pairing, however, the image data is difficult to obtain.

Disclosure of Invention

The present disclosure provides a technical solution for generating a training image.

According to an aspect of the present disclosure, there is provided a method of generating a training image, including:

acquiring an original training image;

randomly selecting a sub-set of the added objects in the added object set;

obtaining a target adding object according to the adding object subset;

and performing image processing on the original training image according to the target addition object to obtain the training image.

In a possible implementation manner, the randomly selecting a subset of the added objects from the set of added objects includes:

randomly generating first reference data;

comparing the first reference data with a preset threshold;

when the first reference data is not larger than the preset threshold value, generating second reference data;

and selecting an added object subset corresponding to the first numerical value interval according to the first numerical value interval in which the second reference data is positioned.

In a possible implementation manner, the randomly selecting a subset of the added objects from the set of added objects further includes:

and when the first reference data is larger than the preset threshold value, taking the empty set as the added object subset.

randomly generating third reference data;

and selecting an added object subset corresponding to the second numerical value interval according to the second numerical value interval in which the third reference data is positioned.

In a possible implementation manner, the obtaining a target added object according to the added object subset includes:

when the adding object subset is a non-empty set, randomly selecting an adding object from the adding object subset as the target adding object;

and under the condition that the subset of the added objects is an empty set, finishing the image processing of the original training image, and taking the original training image as the training image.

In a possible implementation manner, the adding an object according to the target, and performing image processing on the original training image to obtain the training image includes:

correcting the original training image to obtain a corrected training image;

and according to the target addition object, carrying out image processing on the corrected training image to obtain the training image.

In a possible implementation manner, the correcting the original training image to obtain a corrected training image includes:

acquiring a standard image;

extracting feature points of the original training image to obtain at least one feature point, wherein the feature points comprise one or more than two of left-eye pupil feature points, right-eye pupil feature points, nose tip feature points, left mouth corner feature points and right mouth corner feature points;

and performing affine transformation on the original training image according to the feature points and the standard image to obtain a corrected training image.

In a possible implementation manner, the adding an object according to the target, and performing image processing on the corrected training image to obtain the training image includes:

determining a processing mode of the image processing according to the adding object subset to which the target adding object belongs;

and according to the target addition object, carrying out image processing on the corrected training image according to the image processing mode to obtain the training image.

In one possible implementation, the processing manner of the image processing includes:

a pasting mode; and/or the presence of a gas in the gas,

the pixel value is changed.

In one possible implementation manner, in a case that the processing manner of the image processing includes a paste manner, the image processing includes:

pasting the target adding object to a preset position of the correction training image; alternatively, the first and second electrodes may be,

and randomly selecting a position as a target position in the preset position range of the correction training image, and pasting the target adding object to the target position.

In a possible implementation manner, in a case that the processing manner of the image processing includes a pixel value changing manner, the image processing includes:

and randomly selecting a range within the preset position range of the correction training image as a target range, and changing the pixel value of each pixel point of the target range into a preset pixel value.

In one possible implementation, the adding the subset of objects includes:

a first additional object subset obtained by object extraction; and/or the presence of a gas in the gas,

and generating the obtained second additional object subset through the template.

In one possible implementation, the first add-object subset includes:

a hat add object subset; and/or, the mask adds a subset of objects.

In one possible implementation, the second add-object subset includes: a subset of sunglass objects.

In one possible implementation, the method further includes:

obtaining a matched training image according to the original training image and the training image;

and training a preset neural network model according to the matched training images.

According to an aspect of the present disclosure, there is provided a training image generation apparatus including:

the original training image acquisition module is used for acquiring an original training image;

the selecting module is used for randomly selecting a sub-set of the added objects in the added object set;

the target adding object acquisition module is used for acquiring a target adding object according to the adding object subset;

and the training image acquisition module is used for carrying out image processing on the original training image according to the target addition object to obtain the training image.

In one possible implementation, the selecting module is configured to:

randomly generating first reference data;

comparing the first reference data with a preset threshold;

In one possible implementation, the selecting module is further configured to:

In one possible implementation, the selecting module is configured to:

randomly generating third reference data;

In a possible implementation manner, the target added object obtaining module is configured to:

In one possible implementation, the training image acquisition module is configured to:

correcting the original training image to obtain a corrected training image;

In one possible implementation, the training image acquisition module is further configured to:

acquiring a standard image;

a pasting mode; and/or the presence of a gas in the gas,

the pixel value is changed.

In one possible implementation, the adding the subset of objects includes:

In one possible implementation, the first add-object subset includes:

a hat add object subset; and/or, the mask adds a subset of objects.

The apparatus further comprises a training module to:

According to an aspect of the present disclosure, there is provided an electronic device including:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to: the above-described training image generation method is performed.

According to an aspect of the present disclosure, there is provided a computer-readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the above-described training image generation method.

In the embodiment of the disclosure, the added object subset is randomly selected from the added object set, a target added object is obtained according to the added object subset, and the original training image is subjected to image processing according to the target added object, so that the training image is obtained. Through the process, the existing original training image can be utilized, the object is added based on the randomly selected target, so that a large number of training images added with the target addition object can be obtained relatively simply, and the target addition object is obtained based on the randomly selected addition object subset, so that the reliability, the reality and the diversity of the training images can be further improved along with the difference of the types and the number of the addition object subsets.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments, which proceeds with reference to the accompanying drawings.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure.

Fig. 1 shows a flow chart of a method of generating a training image according to an embodiment of the present disclosure.

Fig. 2 is a schematic diagram illustrating a specific process of obtaining an added object according to an embodiment of the present disclosure.

FIG. 3 shows a schematic diagram of a first subset of add objects, according to an embodiment of the present disclosure.

Fig. 4 shows a schematic diagram of an application example according to the present disclosure.

Fig. 5 shows a schematic diagram of an application example according to the present disclosure.

Fig. 6 shows a schematic diagram of an application example according to the present disclosure.

Fig. 7 shows a schematic diagram of an application example according to the present disclosure.

Fig. 8 shows a block diagram of a generation apparatus of a training image according to an embodiment of the present disclosure.

Fig. 9 shows a block diagram of an electronic device in accordance with an embodiment of the disclosure.

FIG. 10 shows a block diagram of an electronic device in accordance with an embodiment of the disclosure.

Detailed Description

Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. In the drawings, like reference numbers can indicate functionally identical or similar elements. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

The term "and/or" herein is merely an association describing an associated object, meaning that three relationships may exist, e.g., a and/or B, may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the term "at least one" herein means any one of a plurality or any combination of at least two of a plurality, for example, including at least one of A, B, C, and may mean including any one or more elements selected from the group consisting of A, B and C.

Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better understanding of the present disclosure. It will be understood by those skilled in the art that the present disclosure may be practiced without some of these specific details. In some instances, methods, means, elements and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the present disclosure.

Fig. 1 shows a flowchart of a method for generating a training image according to an embodiment of the present disclosure, which may be applied to an image processing apparatus, which may be a terminal device, a server, or other processing devices. The terminal device may be a User Equipment (UE), a mobile device, a User terminal, a cellular phone, a cordless phone, a Personal Digital Assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, or the like.

In some possible implementations, the method for generating the training image may be implemented by a processor calling computer readable instructions stored in a memory.

As shown in fig. 1, the method for generating the training image may include:

in step S11, an original training image is acquired.

In step S12, a subset of the added objects is randomly selected from the set of added objects.

And step S13, obtaining the target added object according to the added object subset.

And step S14, performing image processing on the original training image according to the target adding object to obtain a training image.

Randomly selecting an added object subset from the added object set, obtaining a target added object according to the added object subset, and performing image processing on an original training image according to the target added object to obtain a training image. Through the process, the existing original training image can be utilized, the object is added based on the randomly selected target, so that a large number of training images added with the target addition object can be obtained relatively simply, and the target addition object is obtained based on the randomly selected addition object subset, so that the reliability, the reality and the diversity of the training images can be further improved along with the difference of the types and the number of the addition object subsets.

In the above disclosed embodiments, the specific implementation manner of the original training image is not limited. In a possible implementation manner, the original training image may be a face training image or a training image including a face region, and when the original training image is the face training image, the method provided in the embodiment of the present disclosure may be applied to a face recognition process. Of course, the method provided in the embodiment of the present disclosure may also be applied to other identification processes, such as pupil identification, human body identification, plant identification, or animal identification, and the content of the corresponding original training image may also be changed correspondingly with the change of the application type, for example, when being applied to pupil identification, the original training image may be an eye training image, when being applied to human body identification, a human body training image, when being applied to plant identification, when being applied to animal identification, an animal identification image, and the like.

Since there may be various situations in the implementation manner of the original training image, in step S11, the manner of acquiring the original training image may also vary according to the difference of the original training image, and is not limited in the embodiment of the present disclosure. In a possible implementation manner, when the original training image is a face training image, the original training image may be obtained by directly reading the training image from various existing face image databases to serve as the original training image. Similarly, when the original training image is other types of training images, the original training image may be read from a corresponding training image database.

With the change of the original training image and the difference of the application manner, the types of the added objects included in the added object set for adding to the original training image and the specific implementation forms of the added object subset divided according to the types of the added objects also change correspondingly. For example, when the original training image is a human face, the added objects included in the added object set may be objects that block the human face, such as sunglasses, a hat, or a mask; when the original training image is a pupil, the added objects contained in the added object set may affect the objects for identifying the pupil, such as glasses, sunglasses, facial beautification, tears, etc.; when the original training image is a human body, the corresponding added object may be an overcoat, an umbrella, a backpack or the like; when the original training image is a plant, the corresponding added object may be a dew, a building, a bird nest, and the like; when the original training image is an animal, the corresponding added object may be a hair or the like. The implementation manner of the specific divided added object subset can be further flexibly divided according to the types of the added objects, and is not described herein again.

It has been proposed in the foregoing disclosed embodiments that, in one possible implementation manner, the method proposed in the embodiments of the present disclosure may be applied to a process of face recognition. Further, in a possible implementation manner, the method provided in the embodiment of the present disclosure may be applied to recognition of a blocked human face, and therefore, the set of addition objects may include various additives that can block the human face, such as sunglasses, a hat, a mask, a scarf, or a veil, and the like. In a possible implementation manner, common sunglasses, hats, and masks may be used as an implementation manner of adding the object set, in each of the following disclosed embodiments, a generation method of the training image is described in detail by taking an example of obtaining a training image with occlusion of sunglasses, hats, masks, and the like, and other types of obstructions, or a specific process of applying the method to other types of training images may refer to each of the following disclosed embodiments, and is not specifically developed here.

For the above reasons, in one possible implementation, adding the subset of objects may include:

a first additional object subset obtained by object extraction; and/or generating the obtained second additional object subset through the template.

In the above-mentioned embodiments, it has been proposed that the added objects may be sunglasses, caps, masks, etc., however, for these added objects, the implementation manners of some added objects are various, such as caps, masks, etc., there may be more deformation manners in terms of shapes, colors, etc., and the implementation manners of some added objects are single, such as sunglasses, etc., although there are many sunglasses brands or types in the market, due to the shape and color limitations of the sunglasses themselves, when the method is applied to image recognition, the recognition complexity and the discrimination degree are greatly reduced compared to masks or caps, and the recognition difficulty of transparent sunglasses is further reduced. Therefore, for the added objects of caps or masks, it is more likely that different types of caps or masks need to be acquired as the added objects by means of object extraction, and for the added objects of sunglasses, sunglasses can be acquired as the added objects directly by generating a template with a part capable of being flexibly deformed. In the embodiment of the present disclosure, the added objects obtained by the object extraction are taken as a first added object, a set of which is taken as a first added object subset, and the added objects obtained by the template generation are taken as a second added object, a set of which is taken as a second added object subset.

In the embodiment of the present disclosure, the number of the added objects specifically included in the set of all the added object sets and the added object subsets is not limited, and the added objects are flexibly selected according to actual situations. In the above-mentioned embodiment, it is mentioned that the first additional object subset is obtained by object extraction, and the object extraction may include extracting an additional object from an existing image, and a specific implementation manner of the object extraction is not limited in the embodiment of the present disclosure. In one possible implementation, the first subset of added objects may be obtained by feature extraction or other means. In a possible implementation manner, the plurality of first added objects can be obtained to form a first added object subset by matting the picture containing the plurality of first added objects, the matting mode is not limited, matting can be performed through photoshop, and matting can be performed through other special matting tools. In a possible implementation manner, when some face images with shielding are subjected to matting to obtain the shielding objects, since the face images have various differences in positions, sizes and angles of faces, there may be a deviation from a standard face image, and if the matting is directly performed in a manner of matting the shielding objects from the standard face image, the obtained result may be inaccurate. Therefore, in order to solve this problem, fig. 2 shows a schematic diagram of a specific process for obtaining an added object according to an embodiment of the present disclosure, and as can be seen from the diagram, in an example, the human faces with the occlusions may be aligned with parameters such as a follow-up scale, an angle, and a key point of a standard human face image, and then the occluded objects in the corresponding positions are extracted according to the aligned human face image. The alignment may be referred to in the embodiments disclosed later, and not expanded first. Fig. 3 is a schematic diagram illustrating a first add-on object subset according to an embodiment of the present disclosure, and it can be seen from the diagram that in the embodiment of the present disclosure, a plurality of clearer first add-on objects such as caps and masks can be obtained by matting.

Likewise, the second subset of added objects is obtained through template generation, and the template generation may include generating the added objects based on the template, and the manner of template generation is not limited in the embodiment of the present disclosure. In one possible implementation manner, the second added object can be obtained by flexibly deforming according to the main features of the second added object. For example, in the case of an additional object such as sunglasses, when a human face is shielded, a partial area of the human face is mainly shielded by a circular black lens, so that a plurality of black-filled circles with different radii can be generated as templates of the sunglasses to jointly form a second additional object subset.

The added object subset can comprise a first added object subset obtained through object extraction and can also comprise a second added object subset obtained through template generation, the flexibility degree of the added object set can be greatly improved, the number of added objects contained in the added object set is increased, and therefore the convenience and the practicability of the whole training image obtaining process are improved.

Further, due to the fact that the added object subsets obtained in the same manner may have different types of added objects, and when the different types of added objects are combined with the original training image, the combination manner and position may change, at this time, the added object subsets may be further divided according to the types of added objects.

Thus, in one possible implementation, the first add-object subset may include: a hat add object subset; and/or, the mask adds a subset of objects.

In one possible implementation, the second add-on-object subset may include: a subset of sunglass objects.

By further dividing the added object subset according to the specific type of the added object, the specific mode of image processing on the original training image based on the target added object can be conveniently determined according to the type of the subset to which the target added object belongs, so that the reliability and the reality of the obtained training image are improved.

As can be seen from the foregoing disclosure embodiments, the added object set may include one or more added object subsets, and the number and the type of the added object subsets specifically included may be flexibly determined according to the type and the application mode of the original training image, which is not limited in the disclosure embodiments. Therefore, how to randomly select the additional object subset from the additional object set through step S12 can be flexibly determined according to actual situations.

In one possible implementation, step S12 may include:

in step S1211, first reference data is randomly generated.

In step S1212, the first reference data is compared with a preset threshold.

In step S1213, when the first reference data is not greater than the preset threshold, second reference data is generated.

In step S1214, an addition object subset corresponding to the first numerical interval is selected according to the first numerical interval in which the second reference data is located.

In the embodiment disclosed above, by randomly generating the first reference data, comparing the first reference data with the preset threshold, and generating the second reference data when the first reference data is not greater than the preset threshold, the corresponding subset of the added objects is determined according to the first value interval where the second reference data is located, and through this process, the ratio of the target added objects can be controlled by flexibly using the value of the preset threshold, so that a part of the training data set formed by the finally generated training data is related to the target added objects, and the other part is unrelated to the target added objects, thereby greatly improving the reality of the training data, and also enabling a model generated by training based on the training data to have a better recognition result.

In the above-described disclosed embodiment, the first reference data is randomly generated, and a specific generation manner thereof is not limited, and in a possible implementation manner, a random number may be generated by a random program. In order to facilitate the subsequent control of the proportion of the target added object by the preset threshold value, in a possible implementation manner, the generated random number may be a random number within a certain value range, and in an example, a certain random number p may be generated, where p is greater than or equal to 0 and less than or equal to 1.

After the first reference data is generated, whether to generate the second reference data may be determined by comparing the first reference data with a preset threshold. It has been proposed in the above-mentioned embodiments that the specific value of the preset threshold may be used to control the proportion of the target added object, for example, taking the value range of p as [0,1], assuming that the preset threshold is 0.2, when p is less than or equal to 0.2, the second reference data is generated, so as to select the added object subset according to the second reference data, so that when there are enough images included in the original training image set, about 0.2, that is, about 20% of the original training images in these sets are processed, the target object may be selected according to the added object subset, that is, the generated training images have an image proportion related to the target added object of about 20%. Therefore, by changing the specific value of the preset threshold, the proportion of the training image set finally composed of the training images related to the target adding object can be effectively controlled, and the specific value of the preset threshold can be selected by itself, and is not limited to the following disclosed embodiments. In one example, it is empirically known that occlusion is added to 15% of faces in training data, 85% of faces remain unchanged, and the face recognition performance can be effectively improved, so the preset threshold value may be set to 0.15 in the disclosed example.

After the first reference data is compared with the preset threshold, the second reference data may be generated when the first reference data is not greater than the preset threshold, and the implementation manner and the generation process of the second reference data may refer to the first reference data, which is not described herein again, it is noted that the numerical limitation range of the second reference data may be the same as or different from that of the first reference data, and may be flexibly selected, in one example, the second reference data may be denoted as q, and the value range of q is greater than or equal to 0 and less than or equal to 1.

In the above-mentioned embodiment, it has been proposed that the additional object subset may include a plurality of categories according to actual situations, for example, when the occluded face is identified, the additional object subset may include a hat object subset, a mask object subset, a sunglasses object subset, and the like, and therefore, in order to select the additional object, the additional object subset may be randomly selected first, and then the target additional object may be further selected from the additional object subset. In order to avoid the difference between the proportions of different types of added objects in the final set of training images, the data range of the second reference data may be divided into a plurality of intervals on average by referring to the above-mentioned manner of controlling the selection proportion of the target added object by the preset threshold, and the subset of added objects corresponding to the interval may be selected to determine the final target object. Further, the data range may also be divided into intervals of different proportions to change the proportions of different types of added objects, and specifically, which division manner is selected may be selected according to the actual situation, and is not limited to the following disclosed embodiments.

In one possible implementation, the embodiment of the present disclosure divides the second reference data q proposed above into three sections [0,1/3 ], [1/3,2/3 ]) and [2/3,1], so as to achieve an average random selection of three types of added object subsets, namely, a hat, a mask and sunglasses. Specifically, which adding object subset is selected when q falls into which interval, which is not limited in the embodiment of the present disclosure, may be flexibly selected according to actual situations. In one example, when the generated q satisfies 0 ≦ q <1/3, a subset of sunglass add objects may be selected, when 1/3 ≦ q <2/3, a subset of hat add objects may be selected, and when 2/3 ≦ q ≦ 1, a subset of mask add objects may be selected.

Through the above process, the adding object subset can be selected when the first reference data is not greater than the preset threshold, however, when the first reference data is greater than the preset threshold, some irrelevant data can be considered to be added to the original training image to obtain a training image irrelevant to the target adding object; it is also contemplated to use the original training image directly as the final training image, and therefore, in one possible implementation, step S12 may further include:

in step S1215, when the first reference data is greater than the preset threshold, the empty set is taken as the added object subset.

By taking the empty set as the added object subset when the first reference data is greater than the preset threshold, any added object cannot be selected to be added into the original training image when the first reference data is greater than the preset threshold, so that the finally obtained training image is the same as the original training image, the proportion of images without the added objects in the finally obtained training image set can be effectively maintained, and the model obtained based on the training image set has higher recognition accuracy.

The above-mentioned embodiments of the disclosure propose a method for effectively controlling the proportion of images without adding target addition objects and the proportion of images with different types of target addition objects added to a set of training images by generating random numbers twice, but in the practical application process, the two proportions may also be simultaneously controlled by directly generating random numbers once and setting different value intervals. Therefore, in one possible implementation, step S12 may include:

in step S1221, third reference data is randomly generated.

In step S1222, an addition target subset corresponding to the second numerical interval is selected according to the second numerical interval in which the third reference data is located.

The generation process of the third reference data and the implementation manner may refer to the implementation manner of the first reference data, and are not described herein again. In the embodiment of the present disclosure, the numerical range of the third reference data may be the same as or different from that of the first reference data and that of the second reference data, and may be determined flexibly. In one example, the third reference data may be denoted as r, and the value range of r satisfies 0 ≦ r ≦ 1.

As can be seen from the above, in the embodiment of the present disclosure, whether to select an additional object subset and which additional object subset to select may be determined by adjusting the numerical interval ratio of r. In one example, it may be set that when r satisfies 0 ≦ r < 0.05, a sunglasses addition object subset may be selected, when r satisfies 0.05 ≦ r < 0.10, a hat addition object subset may be selected, when r satisfies 0.10 ≦ r ≦ 0.15, a mask addition object subset may be selected, and when r satisfies 0.15 < r ≦ 1, an empty set may be selected as the addition object subset.

By randomly generating third reference data and directly selecting an added object subset corresponding to a second numerical interval according to the second numerical interval in which the third reference data is located, whether a target added object is added or not and the specific type of the target added object during adding can be directly determined by generating a random number once, so that the determination efficiency of the target added object can be further improved, and the efficiency of the whole training image generation process can be improved.

After determining the added object subset, the target added object may be obtained through step S13, and in one possible implementation, step S13 may include:

in step S131, when the subset of addition objects is a non-empty set, an addition object is randomly selected from the subset of addition objects as a target addition object.

In step S132, when the subset of the addition objects is an empty set, the image processing on the original training image is finished, and the original training image is used as the training image.

The following explains the above process by taking the example of adding occlusion to a face image: as can be seen from the above disclosed embodiments, the adding object subset may include a mask object subset, a hat object subset, a sunglasses object subset, and an empty set, and based on the manner in the above process, when the selected adding object subset is a non-empty set such as a mask, a hat, or a sunglasses, one adding object may be randomly selected from the corresponding adding object subset as a target adding object, for example, when the selected adding object subset is a mask adding object subset, one mask may be randomly selected from the set as a target adding object; when the selected additional object subset is the sunglasses additional object subset, since it is proposed in the above-mentioned disclosed embodiment that the sunglasses additional object can be generated by a template, at this time, a sunglasses template can be randomly selected, or a size of the sunglasses template can be randomly selected to be used as the target additional object, and when a size of the sunglasses template is randomly selected, how to add the occlusion is not expanded at this time, which is described in detail in the following disclosed embodiments. When the selected subset of the added objects is an empty set, it may indicate that there is no object in the set that can be selected, and this may indicate that there is no intention to add any object to the original training image, so the process of processing the original training image according to the target added object in step S14 may be skipped, and the current original training image may be directly used as the final training image.

By randomly selecting the adding objects in the subset when the adding object subset is not empty, and directly taking the original training image as the training image when the adding object subset is empty, the obtained training image can be related to any adding object and has higher comprehensiveness and reliability, and the finally obtained training image set can also store the original training image with a certain proportion, so that the model trained based on the training image set has higher recognition effect and recognition accuracy.

After the target addition object is obtained, the original training image may be subjected to image processing in step S14 to obtain a training image. In one possible implementation, step S14 may include:

step S141, correcting the original training image to obtain a corrected training image.

And step S142, carrying out image processing on the corrected training image according to the target addition object to obtain a training image.

The original training image is corrected to obtain the corrected training image, the corrected training image is subjected to image processing according to the target addition object to obtain the training image, the original training image is corrected in the process, so that different processing modes do not need to be adopted due to difference of the image when the image is processed, convenience and efficiency of the whole processing process are improved, and efficiency of the whole training image generation process is improved.

In the above disclosed embodiment, the method for correcting the original training image is not limited, and any correction method that can make the original training image consistent with the standard requirement of image processing can be used as the implementation form of step S141. In one possible implementation, step S141 may include:

in step S1411, a standard image is acquired.

In step S1412, feature point extraction is performed on the original training image to obtain at least one feature point.

And step S1413, performing affine transformation on the original training image according to the feature points and the standard image to obtain a corrected training image.

In the embodiment disclosed above, the specific implementation manner of the standard image may be flexibly determined according to the type of the original training image, for example, if the original training image is a face image, the standard image may be a standard face image in face recognition, if the original training image is a pupil image, the standard image may be a standard eyeball image in pupil recognition, and the like, which are not illustrated herein one by one.

After the standard image is obtained, at least one feature point, the number of the feature points, and the specific manner of feature extraction may be obtained through step S1412, and flexible selection may also be performed according to the specific category of the original training image, which is not limited herein. Still taking face recognition as an example, in an example, the implementation process of step S1412 may be to extract face key points from the original training image through an open source tool such as a Multi-task Convolutional Neural network (MTCNN) or Dlib, and extract corresponding feature points, and in a possible implementation, the feature points may include left-eye pupil feature points (x), and the feature points may include left-eye pupil feature points (x)_le,y_le) Right eye pupil feature point (x)_re,y_re) Nose tip feature point (x)_n,y_n) Left mouth corner characteristic point (x)_lm,y_lm) And right mouth angle feature point (x)_rm,y_rm) And, further, the position (x) of the mouth center feature point can be calculated from the left mouth corner feature point and the right mouth corner feature point_m,y_m) In one example, x may be satisfied_m＝(x_lm+x_rm)/2，y_m＝(y_lm+y_rm)/2。

After the feature points are obtained, the original training image may be affine-transformed by step S1413 according to the feature points and the standard image to obtain a corrected training image. How to perform affine transformation specifically can also be flexibly selected according to the specific category of the original training image, and is not limited to the following disclosed embodiments. Continuing with the above-mentioned face recognition as an example, in an example, the implementation process of step S1413 may be: the position coordinates of the feature points obtained in the above process are used, and specifically, which feature points can be selected according to actual conditions are used, in the example disclosed in the present disclosure, the position coordinates of the left-eye pupil feature point, the right-eye pupil feature point, and the mouth center feature point are selected, coordinates of these points are corrected to be consistent with coordinates of corresponding points in the standard human face through affine transformation, and a corresponding transformation relationship is obtained, so that the original training image is corrected to be consistent with the standard human face, and a function specifically adopted by the affine transformation is not limited in the example disclosed in the present disclosure, for example, an affine transformation function warffine of an OpenCV library can be adopted. Step S1413 is not limited to be implemented by affine transformation, and any transformation relationship between the original training image and the standard face may be determined based on the feature points, so that the original training image is unified into the transformed affine on the standard face, which may be used as an implementation manner of step S1413. These processes can also be used as the implementation of face alignment proposed in the above disclosed embodiments.

In an example, after the correction is performed according to the standard face, each parameter of the obtained corrected training image may satisfy: the width W '178 and the height H' 218, and the left-eye pupil feature point coordinate is (x)_le’,y_le’) The coordinates of the right eye pupil feature point are (x) (70.7,113.0)_re’,y_re’) The coordinates of the characteristic point of the mouth center are (x) 108.23,113.0_m’,y_m’)＝(89.43,153.51)。

The method comprises the steps of extracting feature points of an original training image to obtain the feature points, carrying out affine transformation on the original training image according to the feature points and a standard image to obtain a corrected training image, wherein the process can effectively unify the position of the feature points and the size of the original training image with the position of the standard image, and therefore, when any original training image is processed, a target adding object can be added in the same mode conveniently, and the speed and the efficiency of the whole training image generating process are greatly improved.

After the corrected training image is obtained, a training image may be obtained through step S142. The implementation manner of step S142 can be flexibly changed according to the types of the original training image and the target adding object, and is not limited to the following disclosed embodiments. In one possible implementation manner, step S142 may include:

step S1421, determine a processing mode of image processing according to the adding object subset to which the target adding object belongs.

Step S1422, according to the target addition object, image processing is performed on the corrected training image according to the image processing method, so as to obtain a training image.

The processing mode is determined according to the adding object subset to which the target adding object belongs, and the correction training image is subjected to image processing in combination with the target adding object according to the corresponding processing mode to obtain the training image.

In the above disclosed embodiments, there may be multiple implementation manners for the processing manners, and different target addition objects may also flexibly select corresponding processing manners according to different categories of the target addition objects, and specifically, which processing manner is selected for which category, may be flexibly selected and set, and is not limited to the following disclosed embodiments.

In one possible implementation, the processing manner of the image processing may include: a pasting mode; and/or, a pixel value modification manner.

The pasting mode can be that the target adding object is directly pasted to the correction training image, and the specific pasting mode and pasting position can be flexibly determined according to the actual situation. The pixel value changing mode may be to correspondingly change the pixels at the partial positions of the corrected training image according to the content of the target addition object, and the specific changed pixels and changed pixel values are not limited in the embodiment of the present disclosure, and may be selected according to actual situations. In one possible implementation, the processing of the corrected training image may be implemented by selecting a paste mode for objects belonging to the first subset of added objects, such as caps or masks, obtained by object extraction, since they do not have a uniform shape color standard, and the processing of the corrected training image may be implemented by selecting a pixel value change mode for objects belonging to the second subset of added objects, such as sunglasses, obtained by object generation, since their shape colors are uniform. However, in one possible implementation, the pixel value of the target addition object in the first addition object subset may be changed, the target addition object in the second addition object subset may be pasted, or another combination may be adopted if another processing method is available, and the processing method may be selected as needed.

By setting different processing modes such as pasting or pixel value changing, the image processing process can be more flexible, and the corresponding processing mode can be selected along with the flexible change of the types of the target adding objects, so that the diversity of the training image generation method provided by the embodiment of the disclosure is greatly improved.

In one possible implementation manner, in a case that the processing manner of the image processing includes a paste manner, the image processing may include:

pasting the target adding object to a preset position of the correction training image; or randomly selecting a position as a target position in the preset position range of the correction training image, and pasting the target adding object to the target position.

In the above process, the preset position or the preset position range of the training image is corrected, and the specific coordinate or the coordinate range thereof is not limited in the embodiment of the present disclosure, and can be flexibly determined according to the type of the target adding object and the training image. For example, for the hat adding object, the preset position or the preset position range may be the head or the top of the hair of the face image to the upper side of the eyebrow, for the mask adding object, the preset position or the preset position range may be the mouth or the lower vertex of the nose to the vertex of the chin of the face image, for the sunglasses adding object, the preset position or the preset position range may be the eyes or the left vertex of the left eye to the right vertex of the right eye, and the like, and the setting may be flexible according to the actual situation.

The above process is explained in detail below with respect to objects belonging to the first additional object subset, such as a hat and a mask:

for the hat adding object, the highest point of the left-eye pupil characteristic point and the right-eye pupil characteristic point can be used as a preset position during pasting, then the hat adding object is pasted to the preset position, or the highest point is used as a reference point of a preset position range, 0-20 pixels upwards from the reference point are used as a translation range, a preset position range is obtained, the height degree of different people wearing the hat is simulated, in this way, a position is randomly selected from the preset position range to be used as a target position, and the hat adding object is pasted to the target position. The pasting mode adopted during pasting can be flexibly selected, in one example, a warpAffine function in an OpenCV library of Python or a paste function in a PIL library can be adopted, and the method is not limited herein.

For the mask adding object, the process during pasting is similar to that of a hat adding object, in the application example of the disclosure, the position of the mouth center feature point can be used as a preset position, the mask adding object is pasted to the position, or a translation range formed by the left mouth corner feature point and the right mouth corner feature point is used as a preset position range to simulate the mask deviation condition, a target position is randomly selected in the preset position range, and the mask adding object is pasted to the position. The pasting method is not described in detail herein.

By pasting the target adding object to the preset position of the corrected training image or to the target position randomly selected in the preset position range, the training image related to the target adding object can be obtained relatively simply, and the obtained training image is more real and reliable due to the fact that the position can be changed to a certain degree.

In a possible implementation manner, in a case that the processing manner of the image processing includes a pixel value modification manner, the image processing may include:

and randomly selecting a range within the preset position range of the corrected training image as a target range, and changing the pixel value of each pixel point of the target range into a preset pixel value.

In the above process, the preset position range may refer to the above disclosed embodiment, and is not described herein again.

The above process is explained in detail below with the object, sunglasses, belonging to the second subset of added objects:

for the sunglasses adding object, because it is mostly black, and the largest difference is that the lenses are different in size, the left-eye pupil feature point and the right-eye pupil feature point can be respectively used as the center of a circle, and a certain radius value is selected to draw a circle, and the range covered by the circle is used as the preset position range, where the radius value may be a certain fixed value or a random value within a certain range, and in the embodiment of the present disclosure, the radius value may be set to a random value within 15 to 45 pixels. Therefore, when the pixel value is changed, a radius value can be randomly selected, a target range covered in a circular shape is determined according to the radius value, and the R, G pixel values of the pixel points of the area involved in the target range and the pixel values of the three channels of B are set to be 0, namely the pixel points in the target range are changed into black, so that the effect of the sunglasses on the face is simulated.

For other target adding objects, the specific numerical values of the preset range and the preset pixel value can be correspondingly set according to the difference of the shapes and the colors of the other target adding objects, so that the effect of the other target adding objects on the original training image can be simulated, the corresponding training data can be obtained, and the corresponding training data are not expanded in detail.

It should be noted that, in the above-mentioned disclosed embodiment, it is proposed that, for a second added object such as a sunglass, there may also be a second added object subset, and the second added object subset may store a random radius value, or may also be a different generated template, when it stores a random radius value, the above-mentioned process may be performed to obtain a training image by making the pixel value higher, and when it stores a different template, it may also be considered to paste the template directly into the corrected training image by way of pasting.

The random selection range in the preset position range of the corrected training image is used as the target range, the pixel value of each pixel point in the target range is changed into the preset pixel value, so that objects with commonality are added to the target, the corresponding pixel value is directly changed, the complex process of scratching and pasting is reduced, and the convenience of the generation process of the whole training image is improved.

In one possible implementation manner, the method provided in the embodiment of the present disclosure may further include:

and step S15, obtaining a matched training image according to the original training image and the training image.

And step S16, training a preset neural network model according to the matched training images.

The paired training images may be image pairs, including the original training images, and the training images obtained after the image processing is performed on the basis of the original training images.

After the training images are obtained, the preset neural network model can be naturally trained according to the training images, so that the trained neural networks can be applied to the actual image recognition and other processing processes. Specifically, which neural network models are trained can be flexibly determined according to the types of the original training images and the target adding objects, and no limitation is made here. In a possible implementation manner, the training image obtained in the embodiment of the present disclosure may be applied to face recognition, and the face recognition has a plurality of different neural network models. In one example, for recognizing a face with occlusion, such as a face occluded by a mask, sunglasses, or hat, the original training image and the training image may be paired, so as to facilitate the neural network to distinguish the difference and the corresponding relationship between the images with occlusion and without occlusion.

The original training image and the training image are paired, and the preset neural network model is trained based on the pairing result, so that the generation method of the training image provided by the embodiment of the disclosure can be effectively applied to the process of face recognition with shielding, the robustness of the face recognition model to various shielding is enhanced, and the technical effects of the accuracy and the efficiency of face recognition with shielding are improved.

Application scenario example

Face recognition technology is widely used in security, finance, information, education and other fields. The face recognition technology is based on face key feature extraction and comparison to complete recognition, so the integrity of features has great influence on the accuracy of recognition. With the development of deep learning technology, the accuracy of face recognition under controlled conditions has reached an ideal effect. However, when the human face is shielded (such as sunglasses, a hat and a mask), the image characteristics of the human face are not complete any more, and the performance of the algorithm is obviously reduced. In order to improve the recognition rate of the shielded face, in a possible implementation manner, reconstruction of the shielded face can be considered, so that an unshielded region is enhanced, the method can also improve the performance to a certain extent, but the method has large calculation amount, high complexity and poor effect; in a possible implementation manner, corresponding occlusion paired face data can also be added to effectively improve the recognition performance, however, such data is difficult to acquire.

Therefore, the method for generating the training images containing the shielding in large quantity can effectively improve the accuracy of the recognition of the face containing the shielding.

Fig. 4 to 7 are schematic diagrams illustrating an application example according to the present disclosure, and as shown in the drawings, an embodiment of the present disclosure proposes a generation method of a training image, where a specific process of the generation method may be:

an obstruction target may be acquired first as an added object added to the face image. In the application example of the present disclosure, the manner of obtaining the object of the shielding object may be obtaining through matting, and the specific process may be: inputting an image containing an obstruction, which can be a face image containing the obstruction in the application example of the disclosure; because the face images have different positions, sizes and angles, in order to ensure that the subsequent shielding template can be more truly added to the face, the scales, angles, key points and the like of the face images can be aligned to the standard face, so that the coordinates of key parts can be obtained by using a face key point detection method, and then the shielded face is aligned with the standard face based on the coordinates; after alignment, an image processing tool such as photoshop with a matting function can be used to obtain templates of various shields such as masks and hats, and the templates are used as the added objects in the added object set.

In an application example of the present disclosure, the added object set that can be obtained through the above process may include 2000 mask templates and 2000 hat templates.

After the set of added objects composed of the occlusion objects is obtained, a face training image including the occlusion can be generated according to the added objects, and a specific process is shown in fig. 4, and as can be seen from the figure, when the face training image including the occlusion is generated, a complete process can be as follows: inputting a face image, and acquiring coordinates of a key point part by using the face key point detection method in the process; then aligning the face with a standard face based on the coordinates of the key points; then, generating a random number p, if p is less than or equal to 0.15, adding a shielding template to the aligned face, otherwise, keeping the shielding template unchanged, so that the shielding template can be randomly added to the aligned face according to the proportion of 15%; when p is less than or equal to 0.15, a random number q is generated, when q is more than or equal to 0 and less than 1/3, a sunglasses template is added to the aligned picture, when q is more than or equal to 1/3 and less than 2/3, a hat shielding template is added, and when q is more than or equal to 2/3 and less than or equal to 1, a mask template is added, so that shielding templates of sunglasses, hats and mask types can be randomly added to the face needing shielding in equal proportion.

Specifically, in an application example of the present disclosure, the key point detection process mentioned in the above process may be:

the method for locating the positions of the key points of the face by using an open source method such as MTCNN or an open source tool such as dlib, in an application example of the present disclosure, MTCNN may be used to obtain the left eye pupil (x)_le,y_le) Right eye pupil (x)_re,y_re) Nose tip (x)_n,y_n) Left mouth angle (x)_lm,y_lm) Right mouth angle (x)_rm,y_rm) Calculating to obtain the center of the mouth (x)_m,y_m) I.e. x_m＝(x_lm+x_rm)/2，y_m＝(y_lm+y_rm)/2。

Further, based on the obtained key points, the specific process of performing face correction or face alignment may be:

using (x) obtained in step 1_le,y_le)，(x_re,y_re)，(x_m,y_m) The three points correct the face to a standard face by using an affine transformation method, in an application example of the disclosure, the affine transformation may adopt an affine transformation function warpAffeine of an Opencv library, and in a possible implementation manner, the affine transformation may also be replaced by a similar transformation. The coordinates of the corrected human face width, height, left eye pupil, right eye pupil and mouth center are consistent with those of the standard face and can be respectively marked as W ', H', (x)_le’,y_le’)，(x_re’,y_re’) And (x)_m’,y_m’). In an application example of the present disclosure, the coordinate values of the parameters obtained after transformation may be: 178, 218, x_le’＝70.7，y_le’＝113.0，x_re’＝108.23，y_re’＝113.0，x_m’＝89.43，y_m’＝153.51。

Further, the specific process of adding the occlusion template based on the random numbers p and q may be:

according to experience, 15% of human faces in training data are shielded, 85% of human faces are kept unchanged, and the human face recognition improvement performance is relatively large. Thus, in an example of the application of the present disclosure, a random number p (0. ltoreq. p.ltoreq.1) may be generated, and if p.ltoreq.0.15, a random number q (0. ltoreq. q.ltoreq.1) may be generated:

if q is 0 or less<1/3, a sunglass mask may be added to the corrected face image. Because the template of sunglasses is single, and many transparent sunglasses actually reduce the degree of difficulty that the sunglasses sheltered from discernment, so in this disclosure application example, do not obtain the template through the mode of scratching to the sunglasses, but adopt the sunglasses template based on pixel of artificial simulation. Since sunglasses are mostly black, the biggest difference is that the lenses are different in size, the application examples of the present disclosure are respectively given by (x)_le’,y_le’)，(x_re’,y_re’) Taking the circle as the center, taking R as the radius (R is a random value of 15-45), setting the pixel values of the three channels of R, G and B of the pixel points of the area related to the circle to 0, and the adding effect is shown in FIG. 5.

If 1/3 is not more than q <2/3, a hat occlusion can be added to the corrected face image. When the hat is added for shielding, one hat template can be randomly selected each time, because the template images are aligned, the hat template is randomly translated up and down when the hat template is pasted to the corrected face image, the height degree of different people wearing the hat can be simulated, the translation range can be 0-20 pixels, the pasting process can refer to warpAffene function of python opencv, paste function of a PIL library and the like, and the adding effect is shown in FIG. 6.

If q is greater than or equal to 2/3 and less than or equal to 1, mask shading can be added to the corrected face image. The process of adding mask shielding is similar to that of cap shielding, one mask template can be randomly selected from the obtained mask templates each time, the aligned mask template is directly pasted to the corresponding position of the corrected face image, and the adding effect is shown in fig. 7.

It should be noted that the image processing method according to the embodiment of the present disclosure is not limited to be applied to the above-mentioned generation of a face image including a blocking object, and is not limited to the technical field of face recognition, and may be applied to the generation of any training image, and the present disclosure does not limit this.

It is understood that the above-mentioned method embodiments of the present disclosure can be combined with each other to form a combined embodiment without departing from the logic of the principle, which is limited by the space, and the detailed description of the present disclosure is omitted.

It will be understood by those skilled in the art that in the method of the present invention, the order of writing the steps does not imply a strict order of execution and any limitations on the implementation, and the specific order of execution of the steps should be determined by their function and possible inherent logic.

Fig. 8 shows a block diagram of a generation apparatus of a training image according to an embodiment of the present disclosure. The training image generating device can be a terminal device, a server or other processing devices. The terminal device may be a User Equipment (UE), a mobile device, a user terminal, a cellular phone, a cordless phone, a Personal Digital Assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, or the like.

In some possible implementations, the image processing apparatus may be implemented by a processor calling computer readable instructions stored in a memory.

As shown in fig. 8, the training image generating device 20 may include:

an original training image obtaining module 21, configured to obtain an original training image.

And a selecting module 22, configured to randomly select a subset of the added objects from the added object set.

And the target adding object obtaining module 23 is configured to obtain a target adding object according to the adding object subset.

And the training image acquisition module 24 is configured to perform image processing on the original training image according to the target addition object to obtain a training image.

In one possible implementation, the selection module is configured to: randomly generating first reference data; comparing the first reference data with a preset threshold; when the first reference data is not larger than a preset threshold value, generating second reference data; and selecting an adding object subset corresponding to the first numerical value interval according to the first numerical value interval in which the second reference data is positioned.

In one possible implementation, the selecting module is further configured to: and when the first reference data is larger than the preset threshold value, taking the empty set as an added object subset.

In one possible implementation, the selection module is configured to: randomly generating third reference data; and selecting an adding object subset corresponding to the second numerical value interval according to the second numerical value interval in which the third reference data is positioned.

In one possible implementation manner, the target added object obtaining module is configured to: when the adding object subset is a non-empty set, randomly selecting an adding object from the adding object subset as a target adding object; and under the condition that the subset of the added objects is an empty set, finishing the image processing on the original training image, and taking the original training image as the training image.

In one possible implementation, the training image acquisition module is configured to: correcting the original training image to obtain a corrected training image; and according to the target addition object, carrying out image processing on the corrected training image to obtain a training image.

In one possible implementation, the training image acquisition module is further configured to: acquiring a standard image; extracting feature points of the original training image to obtain at least one feature point, wherein the feature points comprise one or more than two of left-eye pupil feature points, right-eye pupil feature points, nose tip feature points, left mouth corner feature points and right mouth corner feature points; and performing affine transformation on the original training image according to the feature points and the standard image to obtain a corrected training image.

In one possible implementation, the training image acquisition module is further configured to: determining a processing mode of image processing according to the adding object subset to which the target adding object belongs; and according to the target addition object, carrying out image processing on the corrected training image according to the image processing mode to obtain a training image.

In one possible implementation, the processing method of image processing includes: a pasting mode; and/or, a pixel value modification manner.

In one possible implementation manner, in a case where the processing manner of the image processing includes a paste manner, the image processing includes: pasting the target adding object to a preset position of the correction training image; or randomly selecting a position as a target position in the preset position range of the correction training image, and pasting the target adding object to the target position.

In a possible implementation manner, in a case that a processing manner of the image processing includes a pixel value changing manner, the image processing includes: and randomly selecting a range within the preset position range of the corrected training image as a target range, and changing the pixel value of each pixel point of the target range into a preset pixel value.

In one possible implementation, adding the subset of objects includes: a first additional object subset obtained by object extraction; and/or generating the obtained second additional object subset through the template.

In one possible implementation, the first add-object subset includes: a hat add object subset; and/or, the mask adds a subset of objects.

In one possible implementation manner, the apparatus further includes a training module, and the training module is configured to: obtaining a matched training image according to the original training image and the training image; and training a preset neural network model according to the matched training images.

Embodiments of the present disclosure also provide a computer-readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the above-mentioned method. The computer readable storage medium may be a non-volatile computer readable storage medium.

An embodiment of the present disclosure further provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured as the above method.

The electronic device may be provided as a terminal, server, or other form of device.

Fig. 9 is a block diagram of an electronic device 800 in accordance with an embodiment of the disclosure. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, or the like terminal.

Referring to fig. 9, electronic device 800 may include one or more of the following components: processing component 802, memory 804, power component 806, multimedia component 808, audio component 810, input/output (I/O) interface 812, sensor component 814, and communication component 816.

The processing component 802 generally controls overall operation of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing components 802 may include one or more processors 820 to execute instructions to perform all or a portion of the steps of the methods described above. Further, the processing component 802 can include one or more modules that facilitate interaction between the processing component 802 and other components. For example, the processing component 802 can include a multimedia module to facilitate interaction between the multimedia component 808 and the processing component 802.

The memory 804 is configured to store various types of data to support operations at the electronic device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phonebook data, messages, pictures, videos, and so forth. The memory 804 may be implemented by any type or combination of volatile or non-volatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.

The power supply component 806 provides power to the various components of the electronic device 800. The power components 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.

The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and a user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front facing camera and/or a rear facing camera. The front camera and/or the rear camera may receive external multimedia data when the electronic device 800 is in an operation mode, such as a shooting mode or a video mode. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

The audio component 810 is configured to output and/or input audio signals. For example, the audio component 810 includes a Microphone (MIC) configured to receive external audio signals when the electronic device 800 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals may further be stored in the memory 804 or transmitted via the communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

The I/O interface 812 provides an interface between the processing component 802 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.

The sensor assembly 814 includes one or more sensors for providing various aspects of state assessment for the electronic device 800. For example, the sensor assembly 814 may detect an open/closed state of the electronic device 800, the relative positioning of components, such as a display and keypad of the electronic device 800, the sensor assembly 814 may also detect a change in the position of the electronic device 800 or a component of the electronic device 800, the presence or absence of user contact with the electronic device 800, orientation or acceleration/deceleration of the electronic device 800, and a change in the temperature of the electronic device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 may access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

In an exemplary embodiment, the electronic device 800 may be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components for performing the above-described methods.

In an exemplary embodiment, a non-transitory computer-readable storage medium, such as the memory 804, is also provided that includes computer program instructions executable by the processor 820 of the electronic device 800 to perform the above-described methods.

Fig. 10 is a block diagram of an electronic device 1900 according to an embodiment of the disclosure. For example, the electronic device 1900 may be provided as a server. Referring to fig. 10, electronic device 1900 includes a processing component 1922 further including one or more processors and memory resources, represented by memory 1932, for storing instructions, e.g., applications, executable by processing component 1922. The application programs stored in memory 1932 may include one or more modules that each correspond to a set of instructions. Further, the processing component 1922 is configured to execute instructions to perform the above-described method.

The electronic device 1900 may also include a power component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input/output (I/O) interface 1958. The electronic device 1900 may operate based on an operating system stored in memory 1932, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like.

In an exemplary embodiment, a non-transitory computer readable storage medium, such as the memory 1932, is also provided that includes computer program instructions executable by the processing component 1922 of the electronic device 1900 to perform the above-described methods.

The present disclosure may be systems, methods, and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for causing a processor to implement various aspects of the present disclosure.

The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.

The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.

The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer-readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method of generating a training image, comprising:

acquiring an original training image;

randomly selecting a sub-set of the added objects in the added object set;

obtaining a target adding object according to the adding object subset;

2. The method of claim 1, wherein randomly selecting a subset of the added objects from the set of added objects comprises:

randomly generating first reference data;

comparing the first reference data with a preset threshold;

3. The method of claim 2, wherein randomly selecting a subset of the added objects from the set of added objects, further comprises:

4. The method of claim 1, wherein randomly selecting a subset of the added objects from the set of added objects comprises:

randomly generating third reference data;

5. The method according to any one of claims 1 to 4, wherein the obtaining a target added object according to the added object subset comprises:

6. The method according to any one of claims 1 to 5, wherein the adding an object according to the target, and performing image processing on the original training image to obtain the training image comprises:

correcting the original training image to obtain a corrected training image;

7. The method of claim 6, wherein said correcting said original training image to obtain a corrected training image comprises:

acquiring a standard image;

8. An apparatus for generating a training image, comprising:

9. An electronic device, comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to invoke the memory-stored instructions to perform the method of any of claims 1 to 7.

10. A computer readable storage medium having computer program instructions stored thereon, which when executed by a processor implement the method of any one of claims 1 to 7.