KR102668641B1 - Image processing methods and devices, electronic devices, and computer-readable storage media - Google Patents

Abstract

The present invention provides an image processing method and device, an electronic device, and a computer-readable storage medium, wherein a target shielding object having a shielding relationship with a shielded target object is selected based on the shielding relationship between adjacent objects in a target image. Next, based on the modal mask of the selected target object and the modal mask of the masked target object, the contour information of the shielded portion of the masked target object and the image information of each pixel point within the outline of the shielded portion are determined, and the target The shielded part of the object can be restored.

Description

Image processing methods and devices, electronic devices, and computer-readable storage media

Cross-reference to related applications

The present invention was filed based on the Chinese patent application with application number 202010093403.0 and the filing date was February 14, 2020, and claims priority of the Chinese patent application. All contents of the Chinese patent application are incorporated herein by reference. It is quoted.

The present invention relates to the field of computer technology and image processing, and particularly to image processing methods and devices, electronic devices, and computer-readable storage media.

Understanding techniques based on natural scenarios mainly include object detection, instance segmentation, semantic segmentation, etc., and these techniques only consider the area where the object is not occluded, and ignore the occlusion relationship between objects. For example, instance segmentation detects the classification and location of an object in a given image and then segments the region of the visible part of the object. However, since the occlusion relationship between a plurality of objects, the outline of the occluded part of the object, and image information are important components in scenario understanding, ignoring these parts will limit the technical effect of scenario understanding and affect the application effect.

In consideration of this, the present invention provides at least one image processing method and apparatus.

In a first aspect, an embodiment of the present invention provides an image processing method, the method comprising: acquiring a target image; Determining a shielding relationship between adjacent target objects in the target image - modal masks of adjacent target objects are connected to each other; For a target object that is shielded in the target image, based on the shielding relationship, selecting a target shielding object that has a shielding relationship with the shielded target object from the target image; and determining outline information of a shielded portion of the shielded target object and image information of each pixel point within the outline of the shielded portion, based on the modal mask of the target shielded object and the modal mask of the shielded target object. - The modal mask of the target object includes mask information in which the target object is not masked by another target object.

In the first aspect of the present invention, the target shielding object that has a shielding relationship with the shielded target object is selected based on the shielding relationship between adjacent objects in the target image, and then the modal mask of the target shielding object obtained by selection, the shielded object, Based on the modal mask of the target object, the outline information of the shielded part of the target object and the image information of each pixel point within the outline of the shielded part are determined, and the shielded part of the target object can be restored.

In one possible embodiment, determining an occlusion relationship between adjacent target objects in the target image comprises selecting a pair of adjacent objects from the target image, wherein the pair of adjacent target objects includes two adjacent target objects. ; Complementing the outline of one target object in the adjacent object pair by another target object in the adjacent object pair, respectively; and determining a shielding relationship between two adjacent target objects in the pair of adjacent objects, based on the entire outline after complementation and the overall outline before complementation of each target object.

In the above embodiment, based on the overall outline after complementation and the overall outline before supplementation of the target object, the occlusion relationship between two adjacent objects can be relatively accurately determined, thereby improving the accuracy of the determined occlusion relationship.

In one possible embodiment, the step of each complementing a contour occluded by one target object in the pair of adjacent objects by another target object in the pair of adjacent objects is performed by a first neural network.

In the above embodiment, the trained first neural network is used to determine the outline of a specified target object being occluded by another target object in an adjacent object pair, thereby improving the degree of automation, efficiency, and accuracy of image supplementation.

In one possible embodiment, the step of determining contour information of a shielded portion of the shielded target object based on the modal mask of the target shielded object and the modal mask of the shielded target object includes: determining the union of the modal masks of; and determining, using the first neural network, outline information of a occluded portion of the occluded target object according to an image corresponding to the union and a modal mask of the occluded target object.

In the above embodiment, using the trained first neural network, contour supplementation is performed on the occluded portion of the occluded target object based on the merged modal mask of the target occluded object and the modal mask of the occluded target object. , the accuracy of image supplementation can be improved.

In one possible embodiment, the first neural network comprises: acquiring a first sample image corresponding to a modal mask of a first sample object and a second sample image corresponding to a modal mask of a second sample object; and masking a part of the first sample image using the second sample image, and inputting the first sample image, the second sample image, and the masked first sample image into a first neural network to be trained, The first sample image is obtained by training using a step of adjusting network parameters of the first neural network to be trained, with the goal of complementing the outline masked by the second sample image.

In the above embodiment, one sample image uses a occluded portion of another sample image, and the occluded portion of the other object is determined based on the other sample image after occlusion and the other sample image before occlusion. Train a first neural network to determine contour information, and perform contour security on the other masked target as the target of training, thereby ensuring accuracy when the first neural network obtained through training performs image supplementation, At the same time, when training the first neural network, artificially labeled sample images are not used, thereby overcoming the defect of not being able to guarantee accuracy due to artificially labeled sample images.

In one possible embodiment, training the first neural network comprises:

A portion of the second sample image is masked using the first sample image, and the first sample image and the masked second sample image are input to the first neural network to be trained, so that the first sample image The method further includes adjusting network parameters of the first neural network to be trained, with the goal of keeping the outline unchanged.

In the above embodiment, when training the first neural network, the fact that the other sample image partially occludes the sample image is used, and the first neural network is generated based on the sample image after occlusion and the other sample image. By training a neural network, and the goal of training is to keep the outline of the other sample image unchanged, the accuracy of the first neural network obtained through training can be improved when performing image supplementation.

In one possible embodiment, the step of selecting a target occlusion object that has a occlusion relationship with the occluded target object from the target image, based on the occlusion relationship, may include selecting a target occlusion object from the target image based on the occlusion relationship. Selecting at least one level of target shielding object that has a shielding relationship with the shielded target object - the target shielding object of the first level shields the shielded target object, and the target shielding object of the other level is of the previous level. Target shielding - Includes shielding the target.

In the above embodiment, in addition to selecting the target shielding object at the first level that has a direct shielding relationship with the shielded target object, it also selects the target shielding object at the other level that has a shielding relationship with the target shielding object at the previous level. Therefore, it is possible to improve the transparency of the selected target shielding object, which is advantageous in improving the accuracy of image supplementation.

In one possible embodiment, the step of selecting at least one level of target occlusion object that has a occlusion relationship with the occluded target object from the target image may include, based on the occlusion relationship, a occlusion corresponding to the target image. Constructing a directed graph - a node in the shielded directed graph is a target object adjacent to another target object, and an edge in the shielded directed graph indicates a target object that is shielded from one target object in two adjacent target objects - ; And based on the shielded directed graph, determine ancestor nodes of all levels of the node corresponding to the shielded target object, and select the target object corresponding to the ancestor node of each determined level as the target shield of each level of the shielded target object. It includes the steps of using it as an object.

In the above embodiment, a target shielding object that has a shielding relationship with a shielded target object is determined using a shielding directed graph, thereby improving the omnipresence of the target shielding object obtained by screening and improving screening efficiency.

In one possible embodiment, determining the modal mask of the occluded target object, based on the modal mask of the target occluded object, the image information of each pixel point within the outline of the occluded portion of the occluded target object. Determining a non-modal mask of the shielded target object based on the modal mask of the shielded target object and the outline information of the shielded portion of the shielded target object - the non-modal mask of the target object is the target Contains mask information in which the target is not shielded by another target object and mask information in which the target object is shielded by another target object -; determining a union of modal masks of the target shielding object; determining a first intersection of the union and a non-modal mask of the masked target object; A step of shielding a partial target image using an image corresponding to the first intersection - the partial target image is centered on the masked target object in the target image and includes a portion of pixel points around the masked target object. Contains - ; and determining image information of each pixel point within the outline of the occluded portion of the occluded target object, based on the image corresponding to the first intersection and the occluded partial target image.

In the above embodiment, a portion of the partial target image is masked using the intersection of the non-modal mask of the masked target object and the merged modal mask of the target masked object, that is, the partial target image is masked using the masked area of the target object. shield part of the The accuracy of image information of the pixel point determined based on the masked area of the target object and the partial target image after being masked can be improved.

In one possible embodiment, based on the image corresponding to the first intersection and the partial target image after occlusion, image information of each pixel point in the outline of the occluded portion of the occluded target object is determined. The steps are executed by a second neural network.

In the above embodiment, a trained second neural network is used to determine image information of a pixel point in a contour where a specified target object is occluded by another target object in the adjacent target object pair, thereby automating image content supplementation. Improves precision, efficiency and accuracy.

In one possible embodiment, the second neural network includes: acquiring a third sample image corresponding to a modal mask of a third sample object and a fourth sample image corresponding to a modal mask of a fourth sample object; After blocking a portion of the third sample image using the fourth sample image, determining a second intersection of a modal mask of the third sample object and a modal mask of the fourth sample object; And a second neural network to be trained, based on the masked third sample image and the image corresponding to the second intersection, with the goal of complementing the image information of each cell point of the masked portion of the third sample image. was obtained using the steps for adjusting the network parameters.

In the above embodiment, one sample image uses a occluded portion of another sample image and combines it with an image corresponding to the intersection of the modal masks of the two sample objects after occlusion to outline the occluded portion of the target object. Since the second neural network is trained to determine the image information of my pixel point, the accuracy of the second neural network obtained through training can be guaranteed when determining the image information of the pixel point, and at the same time, when training the second neural network , By not using artificially labeled sample images, it overcomes the defect of not being able to guarantee accuracy caused by artificially labeled sample images.

In one possible embodiment, the goal is to supplement the image information of each pixel point of the masked portion of the third sample image, based on the masked third sample image and the image corresponding to the second intersection. Thus, the step of adjusting the network parameters of the second neural network to be trained includes obtaining a fifth sample image corresponding to the third sample object - the fifth sample image is an image centered on the third sample object. and the image includes the third sample object and pixel points around the third sample object; shielding a portion of the fifth sample image using an image corresponding to the second intersection; and a second neural network to be trained, targeting to complement the image information of each pixel point of the masked portion of the third sample image, based on the masked third sample image and the masked fifth sample image. and adjusting network parameters.

In the above embodiment, based on the fifth sample image masked by the image corresponding to the intersection of the modal masks of the two sample objects and the third sample image masked by the image corresponding to one of the two sample objects. The second neural network is trained, and the target of training is to complement the image information of each pixel point of the masked part of the third sample image, so that when the second neural network obtained by training determines the image information of the pixel point, Accuracy can be improved.

In a second aspect, an embodiment of the present invention provides an image processing device, the device comprising: an image acquisition module configured to acquire a target image; A occlusion determination module configured to determine the occlusion relationship between adjacent target objects in the target image - the modal masks of adjacent target objects are connected to each other; an object selection module configured to select a target shielding object that has a shielding relationship with the shielded target object from the target image, based on the shielding relationship, for a shielded target object in the target image; A mask configured to determine, based on the modal mask of the target shielding object and the modal mask of the shielded target object, contour information of a shielded portion of the shielded target object and image information of each pixel point within the outline of the shielded portion. A processing module comprising: a modal mask of a target object comprising mask information where the target object is not masked by other target objects.

In one possible embodiment, the occlusion determination module selects pairs of adjacent objects from the target image, where an adjacent target object includes two adjacent target objects; one target object in the pair of adjacent objects each complements the outline occluded by the other target object in the pair of adjacent objects; and determine a shielding relationship between two adjacent target objects in the pair of adjacent objects, based on the complemented overall outline and the entire outline before supplementation of each target object.

In one possible embodiment, the occlusion determination module is configured to use a first neural network to determine the occlusion when one target object in the adjacent object pair each complements an outline occluded by another target object in the adjacent object pair. One target object in a pair of adjacent objects is configured to complement the outline masked by the other target object in the pair of adjacent objects.

In one possible embodiment, when the mask processing module determines the outline information of the occluded portion of the occluded target object based on the modal mask of the target occluded object and the modal mask of the occluded target object, determine the union of the modal masks of the target occlusion object; It is configured to determine outline information of a occluded portion of the occluded target object according to an image corresponding to the union and a modal mask of the occluded target object, using the first neural network.

In one possible embodiment, the image processing device acquires a first sample image corresponding to a modal mask of a first sample object and a second sample image corresponding to a modal mask of a second sample object; A portion of the first sample image is masked using the second sample image, and the first sample image, the second sample image, and the masked first sample image are input to a first neural network to be trained, It further comprises a network training module for training the first neural network, configured to adjust network parameters of the first neural network to be trained, targeting that the first sample image complements the outline occluded by the second sample image. do.

In one possible embodiment, the mask processing module generates a modal mask of the target occlusion object, an image of each pixel point within the outline of an occluded portion of the occluded target object, based on the modal mask of the occluded target object. When determining information, the mask processing module determines a non-modal mask of the shielded target object based on the modal mask of the shielded target object and the outline information of the shielded portion of the shielded target object, and - target A non-modal mask of an object includes mask information in which the target object is not occluded by another target object and mask information in which the target object is occluded by another target object. determining a union of modal masks of the target shielding object; determine a first intersection of the union and the non-modal mask of the occluded target object; Masking a partial target image using an image corresponding to the first intersection, wherein the partial target image is centered on the masked target object in the target image and includes a portion of pixel points around the masked target object. Ham - ; Based on the image corresponding to the first intersection and the partial target image after being masked, it is configured to determine image information of each pixel point within the outline of the masked portion of the masked target object.

In a third aspect, the present invention provides an electronic device, the electronic device includes a processor, a memory, and a bus, the memory stores machine-readable instructions executable by the processor, and the electronic device operates. When enabled, the processor and the memory communicate via a bus, and when the machine-readable instructions are executed by the processor, steps of the image processing method are executed.

In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is operated by a processor, the image processing method steps are executed. do.

Since the devices, electronic devices, and computer-readable storage media of the present invention include technical features that are substantially the same as or similar to those of at least one aspect or embodiment of any one aspect of the method of the present invention, The effect description regarding the apparatus, electronic device and computer-readable storage medium may refer to the effect description in the method content, and the description will not be repeated here.

In order to more clearly explain the technical solutions of the embodiments of the present invention, the accompanying drawings that should be used in the embodiments are briefly introduced below. The accompanying drawings below only illustrate some embodiments of the present invention and are therefore limited to the scope. It should not be considered limiting, and it should be understood that a person skilled in the art may obtain other related accompanying drawings in accordance with these accompanying drawings, even under the premise of not assigning creative labor.
1 is a flowchart of an image processing method provided in an embodiment of the present invention.
Figure 2 is a flowchart for determining a shielding relationship in another image processing method provided by an embodiment of the present invention.
Figure 3 is an example structure of a shielded directed graph in another image processing method provided in an embodiment of the present invention.
Figure 4 is a flowchart of determining image information of each pixel point within the outline of the occluded portion of the occluded target object in another image processing method provided by an embodiment of the present invention.
Figure 5 is a flowchart of training a second neural network in another image processing method provided in an embodiment of the present invention.
Figure 6 is an example flow diagram of training a first neural network in another image processing method provided in an embodiment of the present invention.
Figure 7 is an example flow diagram of training a second neural network in another image processing method provided in an embodiment of the present invention.
Figure 8 is an exemplary structural diagram of an image processing device provided in an embodiment of the present invention.
Figure 9 is an exemplary structural diagram of an electronic device provided in an embodiment of the present invention.

In order to further clarify the purpose, technical solutions and advantages of the present invention, below, in combination with the drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are clearly and completely described. , the drawings of the present invention are for illustrative and descriptive purposes only and are not used to limit the scope of protection of the present invention. Additionally, it should be understood that the example drawings are not drawn to scale. The flowcharts used in the present invention are illustrated according to the operations implemented in some embodiments of the present invention. It should be understood that operations in a flowchart may not be implemented in order, and steps for which the upper and lower contexts are not logical may be implemented in reverse order or simultaneously. Additionally, with the guidance of this disclosure, one skilled in the art may add one or more other operations to the flowchart or remove one or more operations from the flowchart.

Additionally, the described embodiments are only some embodiments disclosed in the present invention and are not all embodiments. Components of embodiments of the invention generally described and shown in the drawings herein may be arranged and designed in a variety of different configurations. Accordingly, the detailed description of embodiments of the present invention provided in the drawings below are not intended to limit the scope of protection of the present invention, but are intended to indicate selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative labor shall fall within the protection scope of the present invention.

It should be noted that, in embodiments of the present invention, the term "included" is used to indicate the presence of a subsequently declared feature, but does not exclude other features from being added.

In the technology of natural scenario understanding, the occlusion relationship between a determined plurality of objects, the contour of the occluded part of the object and the method for determining and improving the technical tasks of image information, the present invention relates to image processing methods and devices, electronic devices, computer reading Provide available storage media. Here, the present invention selects target shielding objects that have a shielding relationship with the shielded target object based on the shielding relationship between adjacent target objects in the target image, and then selects the modal mask of the target shielding object obtained by screening, and the mask of the shielded target object. Based on the modal mask, the outline information of the masked part of the masked target object and the image information of each pixel point within the outline of the shielded part are determined, and the shielded part of the target object can be restored.

Below, image processing methods and devices, electronic devices, and computer-readable storage media will be described through specific embodiments.

An embodiment of the present invention provides an image processing method, and the method is applied to a terminal device that performs compensation for a occluded target object. Specifically, as shown in Figure 1, the image processing method provided in the embodiment of the present invention includes the following steps.

In step S110, a target image is acquired.

Here, the target image includes at least one masked target object. The target image may be captured by the terminal device using its own camera, or may be received by the terminal device from another device, and the present invention is not limited to the device that captures the target image.

In step S120, a occlusion relationship between adjacent target objects in the target image is determined.

Here, the modal masks of adjacent target objects are connected to each other. First, all adjacent target objects are selected from the target image. An adjacent target object can be understood as a target object with which a shielding relationship exists. After selecting and obtaining adjacent target objects, the shielding relationship between the two target objects is determined.

In some possible implementations, existing image processing methods can be used to select targets that are occluded from each other.

In some possible implementation methods, the occluded target object may first be supplemented and then combined with the supplemented target object to determine the occlusion relationship.

In step S130, for a target object that is shielded in the target image, a target shielding object that has a shielding relationship with the shielded target object is selected from the target image based on the shielding relationship.

Based on the determined shielding relationship, a corresponding target shielding object is selected for the shielded target object, and the target shielding object not only includes a target object that directly shields the shielded target object, but also includes the shielded target object. It further includes a target object that directly shields the target object, that is, the target shield object can be divided into different levels of shielding objects, and the target shielding object of the first level is a target object that directly shields the shielded target object. , and the target shielding object of another level shields the target shielding object of the previous level.

Here, it should be explained that the selected target shielding object includes at least one level of shielding object.

In step S140, based on the modal mask of the target occlusion object, the modal mask of the occluded target object, contour information of the occluded portion of the occluded target object, and image information of each pixel point in the outline of the occluded portion. Decide.

Here, the modal mask of the target object includes mask information that the target object is not shielded by other target objects. Before determining the occluded contour information of the target object and the image information of the pixel points in the occluded outline, first determine the modal mask of the occluded target object and the modal mask of the target occluded object, and then based on each determined modal mask, Contour information and image information of pixel points must be determined.

The modal mask includes mask information of the unshielded part of the shielded target object, that is, information of the object to which each pixel point corresponding to the unshielded part of the shielded target object belongs, and specifically, the mask information of the unshielded part of the shielded target object. Each pixel point corresponding to the unoccluded portion may include an identification ID of the object to which it belongs. In some possible implementation methods, using a pre-trained neural network, based on the modal mask of the target occlusion object, the modal mask of the occluded target object, contour information of the occluded portion of the occluded target object, and the occlusion of the occluded portion. Image information for each pixel point within the outline can be determined.

In the above embodiment, the accuracy of the outline information and image information of the determined shielding portion can be effectively improved by combining the target object with the target shielding object.

In some embodiments, as shown in Figure 2, the occlusion relationship between adjacent target objects in the target image is determined, and is specifically implemented using the steps below.

In step S210, adjacent object pairs are selected from the target image.

Here, one adjacent target pair includes two adjacent target objects. Adjacent target objects in the target image are determined, and if there are more than two determined adjacent target objects, for example, if each target object in the three target objects is determined to be adjacent to two other target objects, the three targets By combining any two target objects among the objects, a total of three adjacent object pairs must be obtained.

In step S220, one target object in the pair of adjacent objects respectively complements the outline masked by the other target object in the pair of adjacent objects.

Here, processing is performed on each target object in the adjacent object pair using a pre-trained first neural network, so that each target object in the adjacent object pair is masked by another target object in the adjacent object pair. You can get an outline.

In some possible implementation methods, object supplementation is performed by inputting the image corresponding to the modal mask of each target object in the adjacent object pair into the trained first neural network.

Using the trained first neural network, the specified target object determines the outline occluded by another target object in an adjacent object pair, thereby improving the degree of automation, efficiency, and accuracy of image outline complementation.

What needs to be explained is that in this step, when supplementing the object (i.e. determining the occluded outline of the object), not only can the occlusion be performed on the occluded object, but also the unoccluded object can be supplemented. may; The difference is that when supplementing a occluded object, the outline of the occluded part is determined, that is, the determined outline is not empty, and when supplementation is performed on an unoccluded object, it is determined that no outline is obtained, In other words, the determined outline is empty, and the original outline of the object is maintained so as not to change.

In step S220, a shielding relationship between two adjacent target objects in the adjacent object pair is determined based on the entire outline after complementation and the entire outline before complementation of each target object.

In some possible implementations, this step determines the amount by which the area of the target object is increased, based on the post-augmented overall contour and the pre-augmented overall contour of the target object, and then determining the amount by which the area of the two adjacent target objects is increased. By comparing the increased amounts, it can be determined that a target object with a relatively large area increase amount is shielded by a target object with a relatively small area increase amount, that is, a shielding relationship between two adjacent target objects can be determined.

In the above embodiment, based on the overall outline after complementation and the overall outline before supplementation of the target object, the shielding relationship between two adjacent target objects can be determined relatively accurately, improving the accuracy of the determined shielding relationship.

In related technologies, when performing contour supplementation for a target object, a neural network is generally first trained using an artificially labeled sample image, and then the contour of the target object is supplemented using the trained neural network. Artificially labeling sample images not only wastes human resources, but also affects the detection accuracy of the trained neural network by introducing low labeling accuracy defects due to the uncertainty of artificial experience.

Regarding the above technical problem, the present invention provides the following embodiment to avoid training a neural network configured to perform contour complementation on a target object using artificially labeled sample images.

Obtaining a first sample image corresponding to the modal mask of the first sample object and a second sample image corresponding to the modal mask of the second sample object; A portion of the first sample image is masked using the second sample image, and the first sample image, the second sample image, and the masked first sample image are input to a first neural network to be trained, The network parameters of the first neural network to be trained are adjusted with the goal that the first sample image complements the outline masked by the second sample image.

The above may mean shielding part of the first sample image using the second sample image, or erasing part of the first sample image using the second sample image.

In the above embodiment, the target of training is to perform contour supplementation on the masked first sample image, that is, to restore the masked first sample image to the first sample image before masking, so that the first neural network obtained by training Accuracy can be guaranteed when performing this contour complementation. At the same time, when training the first neural network in the above embodiment, artificially labeled sample images are not used, thereby overcoming the defect of not being able to guarantee accuracy due to artificially labeled sample images.

The first neural network obtained by training in the above embodiment can complement the outline of the masked target object, and for the unmasked object, when the first neural network proceeds with processing, the original outline must be kept unchanged. Therefore, the step of training the first neural network must further include the following steps.

A portion of the second sample image is masked using the first sample image, and the first sample image and the masked second sample image are input to a first neural network to be trained, thereby defining the outline of the first sample image. With the goal of keeping this unchanged, the network parameters of the first neural network to be trained are adjusted.

In the above content, the target of training is not to perform contour supplementation on the first sample image, that is, to keep the original contour of the first sample image unchanged, by using the first neural network obtained by training in the above step. , the accuracy when performing image contour supplementation can be further improved.

In some possible implementations, as shown in FIG. 6, the first sample image is A, the second sample image is B, and the second sample image is the masked first sample after masking the first sample image. The image is marked as A\B, and the first sample image, the second sample image, and the first sample image after masking are input to the first neural network to be trained (PCNet-M), and the first sample image after masking is input. The network parameters of the first neural network are adjusted with the goal of restoring to the first sample image before shielding.

At the same time, as shown in FIG. 6, after the first sample image masks the second sample image, the masked second sample image is displayed as B\A, and the second sample image after masking and the first sample image are displayed as B\A. The sample image is input into the first neural network to be trained (PCNet-M), and the network parameters of the first neural network are adjusted with the goal of keeping the outline of the first sample image unchanged.

In some embodiments, the step of selecting a target shielding object that has a shielding relationship with the shielded target object from the target image based on the shielding relationship may be specifically implemented using the following steps.

Based on the shielding relationship, at least one level of target shielding object having a shielding relationship with the shielded target object is selected from the target image, where the target shielding object of the first level is shielded from the shielded target object. And, the target shielding object of another level shields the target shielding object of the previous level.

In the above embodiment, in addition to selecting the target shielding object at the first level that has a direct shielding relationship with the shielded target object, it also selects the target shielding object at the other level that has a shielding relationship with the target shielding object at the previous level. Therefore, it is possible to improve the transparency of the selected target shielding object, which is advantageous in improving the accuracy of image supplementation.

The step of selecting at least one level of target shielding object that has a shielding relationship with the shielded target object from the target image may be specifically implemented using the following steps.

Based on the determined occlusion relationship, construct an occlusion directed graph corresponding to the target image; A node in the occluded directed graph is a target object that is adjacent to another target object, and an edge in the occluded directed graph indicates a target object that is occluded from a target object in two adjacent target objects; Based on the shielded directed graph, ancestor nodes of all levels of nodes corresponding to the shielded target object are determined, and target objects corresponding to the ancestor nodes of each determined level are designated as target shielded objects of each level of the shielded target object. Use it as

In some possible implementations, the constructed shielded directed graph is as shown in Figure 3, where node 5 is shielded by node 2, node 1 is shielded by node 5, node 2, node 4, and node 3. is shielded by If the target is node 1, the determined ancestor nodes of node 1 include node 5, node 2, node 4, and node 3. If the target object is node 7, the determined ancestor nodes of node 7 include node 1, node 3, node 5, node 2, and node 4.

Additionally, the constructed shielded directed graph may also include nodes for which no shielding relationship exists with other nodes, for example nodes other than node 8 and node 9 have no shielding relationship.

In the above embodiment, the target shielding object that has a shielding relationship with the target object is determined using a shielding directed graph, thereby improving the omnipresence of the target shielding object obtained by screening and improving screening efficiency.

In some embodiments, the step of determining the outline information of the shielding portion of the shielded target object based on the modal mask of the target shielded object and the modal mask of the shielded target object, specifically using the following steps: It can be implemented.

determining a union of modal masks of the target shielding object; Using the first neural network, outline information of the occluded portion of the occluded target object is determined according to the image corresponding to the union and the modal mask of the occluded target object.

The first neural network is used to perform contour supplementation for the occluded object based on the modal mask of the object, that is, it determines the outline information of the occluded portion of the occluded object. Therefore, here, using the first neural network, outline information of the shielded portion of the target object can be determined based on the modal mask of the target shielded object corresponding to the target object and the modal mask of the shielded target object.

Using the first neural network trained in the above embodiment, compensation is performed on the occluded part of the target object based on the merged modal mask of the target occlusion object and the modal mask of the occluded target object, thereby improving the accuracy of image supplementation. can be improved.

After obtaining the outline information of the masked portion of the target object, the non-modal mask of the target object can be determined based on the modal mask of the target object. Here, the non-modal mask is obtained after merging the modal mask of the target object and the outline information of the occluded part of the target object, and not only includes the outline of the unoccluded part of the target object, but also contains the outline of the occluded part of the target object. It also includes outlines.

In some embodiments, as shown in Figure 4, a modal mask of the target occlusion object, an image of each pixel point in the outline of the occlusion portion of the occlusion target object, based on the modal mask of the occlusion target object. The step of determining information can be specifically implemented using the steps below.

In step S410, a non-modal mask of the shielded target object is determined based on the modal mask of the shielded target object and outline information of the shielded portion of the shielded target object.

Here, the non-modal mask of the target object includes mask information in which the target object is not shielded by another target object and mask information in which the target object is shielded by another target object. Specifically, a non-modal mask of the target object can be obtained by merging the modal mask of the shielded target object and the outline information of the shielded portion of the shielded target object. The obtained non-modal mask contains the outline of the unoccluded part of the target object and also contains the outline of the occluded part of the target object.

In step S420, a union of modal masks of the target shielding object is determined; A first intersection of the union and the non-modal mask of the shielded target object is determined.

Here, the first intersection determined using the non-modal mask of the shielded target object and the union is mask information corresponding to each pixel point of the shielded area of the shielded target object.

In step S430, the partial target image is shielded using the image corresponding to the first intersection.

Here, the partial target image centers on the occluded target object in the target image and includes a portion of pixel points around the occluded target object. A partial target image is shielded using the shielded area of the shielded target object.

In step S440, image information of each pixel point within the outline of the masked portion of the masked target object is determined based on the image corresponding to the first intersection and the partial target image after masking.

Here, the image corresponding to the first intersection and the partial target image after being masked are input to a pre-trained second neural network, and the trained second neural network processes the input image, thereby blocking the masking. This may be to obtain image information of each pixel point within the outline of the masked portion of the target object.

This step uses a trained second neural network to determine the image information of pixel points in the outline where a specified target object is occluded by another target object in an adjacent object pair, thereby improving the degree of automation, efficiency, and accuracy of image supplementation. It can be improved.

In the above embodiment, the accuracy of image information of the determined pixel point can be improved by using the masked area of the masked target object and the subsequent masked image corresponding to the target object.

In related technology, when determining image information of a pixel point, a neural network is generally first trained using an artificially labeled sample image, and then the image information of the pixel point is determined using the trained neural network. Artificially labeling sample images not only wastes human resources, but also introduces defects such as low labeling accuracy due to the uncertainty of artificial experience, thereby affecting the detection accuracy of the trained neural network.

Regarding the above technical problem, the present invention provides the following embodiment to avoid training a neural network using artificially labeled sample images, as shown in Figure 5.

In step S510, a third sample image corresponding to the modal mask of the third sample object and a fourth sample image corresponding to the modal mask of the fourth sample object are acquired.

In step S520, a portion of the third sample image is shielded using the fourth sample image.

Here, not only may some of the third sample images be shielded using the fourth sample image, but also some of the third sample images may be erased using the fourth sample image.

In step S530, after blocking a part of the third sample image using the fourth sample image, a second intersection of the modal mask of the third sample object and the modal mask of the fourth sample object is determined.

In step S540, based on the masked third sample image and the image corresponding to the second intersection, the target is to complement the image information of each pixel point of the masked portion of the third sample image, and a second to be trained Adjust the network parameters of the neural network.

In some possible implementation methods, this step may be implemented using the substeps below.

In substep 1, a fifth sample image corresponding to the third sample object is acquired, the fifth sample image is an image centered on the third sample object, and the image is the third sample object and the third sample object. Includes pixel points around the third sample object.

In substep 2, a part of the fifth sample image is masked using the image corresponding to the second intersection.

In sub-step 3, based on the masked third sample image and the masked fifth sample image, training is performed with the goal of complementing the image information of each pixel point of the masked portion of the third sample image. Adjust the network parameters of the second neural network to be used.

In this step, a second neural network is trained using the fifth sample image masked by the corresponding image and the third sample image masked by the image corresponding to one sample object, and the training target is the third sample image. By supplementing the image information of each pixel point in the masked portion of the image, the accuracy of the second neural network obtained through training can be improved when determining the image information of the pixel point.

What should be explained is that the second intersection is the part where the third sample image is erased or blocked by the fourth sample image, and the fifth sample image is masked by the second intersection, but the fifth sample image is substantially masked, The third sample image is the part erased by the fourth sample image, and therefore, when training the second neural network, restoring the fifth sample image can be considered as the target, and the masked part of the third sample image Complementing the image information of each pixel point may be considered a target.

Here, the role of the masked third sample image is to guide the neural network that the object to be restored is the object corresponding to the third sample image.

In the above embodiment, one sample image partially occludes another sample image, and the outline of the occluded portion of the object is combined with the image corresponding to the intersection of the modal masks of the two sample objects after occlusion. By training a second neural network to determine image information of my pixel point, the accuracy of the trained second neural network in determining the information of the pixel point can be guaranteed; At the same time, when training the second neural network, artificially labeled sample images are not used, thereby overcoming the defect of not being able to guarantee accuracy due to artificially labeled sample images.

In some possible implementation methods, as shown in FIG. 7, the third sample image may use the previous first sample image A, the fourth sample image may use the previous second sample image B, and the third sample image may use the previous second sample image B, and After the 4 sample image masks the third sample image, the masked third sample image is displayed as A\B. After masking, the second intersection of the modal mask of the third sample object and the modal mask of the fourth sample object can be expressed as A∩B, and the masked third sample image and the masked fifth sample image It is input to the second neural network to be trained (PCNet-C), and the network parameters of the second neural network are adjusted with the goal of restoring the fifth sample image after masking to the fifth sample image before masking.

It should be noted that the step of training the first neural network and the second neural network may be performed by a terminal device executing the image processing method, or may be performed by a device independently training one or two neural networks.

The image processing method in the above embodiment does not use sample images of artificially labeled occlusion relationships, outline information of the occluded portion of the target object, and image information of pixel points within the occluded outline, thereby saving human resources. , overcoming the uncertainty caused by artificial labeling.

In the above embodiment, the target object masked in the target image can be supplemented using the trained first neural network, and the occlusion relationship between adjacent target objects in the target image is predicted based on the outline before and after the target object complementation. do. In the above embodiment, the target shielding object corresponding to the shielded target object can be determined based on the shielding relationship, thereby determining the modal mask of the determined target shielding object and the shielding of the shielded target object based on the modal mask of the shielded target object. Determine the outline information of the part and the image information of each pixel point within the outline of the masked part. In the above embodiment, the network trained using artificially labeled images is prevented from determining the outline information and image information of the occluded portion, and at the same time, the target occluded object of the occluded target object is connected to the occluded portion of the occluded target object. By combining this with supplementation, the accuracy of the contour information and image information of the determined shielded part can be effectively improved.

Corresponding to the above image processing method, the present invention further provides an image processing device, the device being applied to a terminal device that performs compensation for a occluded object, each module capable of implementing the same method steps in the method, and the same beneficial effects can be obtained, and therefore, for the same parts, the present invention will not be described again repeatedly.

Specifically, as shown in FIG. 8, the present invention provides an image processing device, the device comprising:

An image acquisition module 810 configured to acquire a target image;

A shielding determination module 820 configured to determine a shielding relationship between adjacent target objects in the target image - modal masks of adjacent target objects are connected to each other;

An object selection module 830 configured to select a target shielding object that has a shielding relationship with the shielded target object from the target image, based on the shielding relationship, with respect to a shielded target object in the target image; and

A mask configured to determine, based on the modal mask of the target shielding object and the modal mask of the shielded target object, contour information of a shielded portion of the shielded target object and image information of each pixel point within the outline of the shielded portion. Processing module 840 includes a modal mask of a target object including mask information where the target object is not occluded by other target objects.

In some embodiments, the shielding determination module 820:

Select adjacent object pairs from the target image - an adjacent object pair includes two adjacent target objects;

one target object in the pair of adjacent objects each complements an outline masked by another target object in the pair of adjacent objects;

and determine a shielding relationship between two adjacent target objects in the pair of adjacent objects, based on the overall outline after complementation and the overall outline before complementation of each target object.

In some embodiments, the occlusion determination module 820 determines the occlusion when one target object in the adjacent object pair respectively complements the outline occluded by the other target object in the adjacent object pair. )silver,

Using the first neural network, one target object in the adjacent object pair is configured to respectively complement the outline masked by the other target object in the adjacent object pair.

In some embodiments, when the mask processing module 840 determines the outline information of the shielding portion of the shielded target object based on the modal mask of the target shielded object and the modal mask of the shielded target object, The mask processing module 840 is,

determining a union of modal masks of the target shielding object;

It is configured to determine outline information of a occluded portion of the occluded target object according to an image corresponding to the union and a modal mask of the occluded target object, using the first neural network.

In some embodiments, the image processing device further includes a network training module 850 for training the first neural network, and the network training module 850 includes,

Obtaining a first sample image corresponding to the modal mask of the first sample object and a second sample image corresponding to the modal mask of the second sample object;

A portion of the first sample image is masked using the second sample image, and the first sample image, the second sample image, and the masked first sample image are input to a first neural network to be trained, and adjust network parameters of the first neural network to be trained, with the goal that the first sample image complements the outline masked by the second sample image.

In some embodiments, the network training module 850 also masks a portion of the second sample image using the first sample image, and uses the first sample image and the second sample image after the masking. It is configured to adjust network parameters of the first neural network to be trained by inputting it to the first neural network to be trained, with the goal of keeping the outline of the first sample image unchanged.

In some embodiments, the object selection module 830 selects a target occlusion object of at least one level that has a occlusion relationship with the occluded target object from the target image, based on the occlusion relationship - the first level. The target shielding object is configured to shield the shielded target object, and the target shielding object at another level is configured to shield the target shielding object at the previous level.

In some embodiments, the target selection module 830 also constructs, based on the occlusion relationship, a occlusion directed graph corresponding to the target image, wherein a node in the occlusion directed graph is a target object that is adjacent to another target object. , and an edge in the shielded directed graph indicates a target object that is shielded from one target object in two adjacent target objects -; Based on the occluded directed graph, determine ancestor nodes of all levels of the node corresponding to the occluded target object; It is configured to use the target object corresponding to the ancestor node of each determined level as the target shielding object of each level of the shielded target object.

In some embodiments, the mask processing module 840 generates an image of each pixel point within the outline of the occluded portion of the occluded target object based on the modal mask of the target occluded object and the modal mask of the occluded target object. When determining information, the mask processing module 840

Based on the modal mask of the shielded target object and the outline information of the shielded portion of the shielded target object, a non-modal mask of the shielded target object is determined - the non-modal mask of the target object is determined when the target object is different from the target object. Contains mask information that is not shielded by a target object and mask information that the target object is shielded by another target object -;

determining a union of modal masks of the target shielding object;

determine a first intersection of the union and the non-modal mask of the occluded target object;

Masking a partial target image using an image corresponding to the first intersection, wherein the partial target image is centered on the masked target object in the target image and includes a portion of pixel points around the masked target object. Ham - ;

Based on the image corresponding to the first intersection and the partial target image after being masked, it is configured to determine image information of each pixel point within the outline of the masked portion of the masked target object.

In some embodiments, the network training module 850 is also configured to train a second neural network, wherein the second neural network further includes an image corresponding to the first intersection and the portion after being masked. Based on the target image, the image information of each pixel point within the outline of the masked portion of the masked target object is determined.

In some embodiments, the network training module 850 further acquires a third sample image corresponding to the modal mask of the third sample object and a fourth sample image corresponding to the modal mask of the fourth sample object; After blocking a portion of the third sample image using the fourth sample image, determining a second intersection of the modal mask of the third sample object and the modal mask of the fourth sample object; Based on the masked third sample image and the image corresponding to the second intersection, the target is to complement the image information of each pixel point of the masked portion of the third sample image, and the second neural network to be trained It is configured to adjust network parameters.

In some embodiments, the network training module 850 also obtains a fifth sample image corresponding to the third sample object, wherein the fifth sample image is an image centered on the third sample object, and , including the third sample object and pixel points around the third sample object; shielding a portion of the fifth sample image using an image corresponding to the second intersection; A network of a second neural network to be trained, based on the masked third sample image and the masked fifth sample image, with the goal of complementing the image information of each pixel point of the masked portion of the third sample image. It is configured to adjust parameters.

An embodiment of the present invention discloses an electronic device, as shown in FIG. 9, including a processor 901, a memory 902, and a bus 903, wherein the memory 902 includes the processor 901. Executable machine-readable instructions are stored, and when the electronic device is operated, the processor 901 and the memory 902 communicate through a bus 903.

When the machine-readable instruction is executed by the processor 901,

Obtaining a target image;

Determining a shielding relationship between adjacent target objects in the target image - modal masks of adjacent target objects are connected to each other;

For a target object that is shielded in the target image, based on the shielding relationship, selecting a target shielding object that has a shielding relationship with the shielded target object from the target image; and

Based on the modal mask of the target shielding object and the modal mask of the shielded target object, determining outline information of a shielded portion of the shielded target object and image information of each pixel point within the outline of the shielded portion - The modal mask of the target object includes mask information in which the target object is not masked by other target objects.

In addition, when the machine-readable instructions are executed by the processor 901, the method content in any one embodiment described in the method section above is executed, and the description is not repeated here.

An embodiment of the present invention further provides a computer program product corresponding to the method and device, and includes a computer-readable storage medium on which program code is recorded, and instructions included in the program code execute the method of the previous method embodiment. may be used, and specific implementation may refer to the above method embodiments, and will not be repeated herein.

As described above, for each embodiment, we will focus on the differences between each embodiment, and the same or similar parts may be referenced to each other, and for the sake of brevity, they will not be repeated further in the text.

Those skilled in the art may understand that, for convenience and brevity of explanation, the specific operation process of the above-described system and device may refer to the corresponding process in the method embodiments, and may be repeated repeatedly in the present invention. Doesn't explain. It should be understood that, in some embodiments provided herein, the disclosed systems, devices, and methods may be implemented in other ways. The above-described device embodiments are only schematic, for example, the division of the module is only a logical function division, and in actual implementation, there may be other division methods, for example, a plurality of modules or components. It may be combined or integrated into other systems, or some features may be ignored or not implemented. Additionally, the mutual coupling or direct coupling or communication connection shown or discussed is implemented through some communication interface, and the indirect coupling or communication connection through a device or module may be electrical, mechanical or other form.

Modules described as separate members may or may not be physically separated, and members represented as modules may or may not be physical units, that is, they may be located in one place or distributed over a plurality of network units. Depending on actual needs, some or all of the units may be selected to implement the purpose of the method of this embodiment.

Additionally, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may be an independent physical entity, or two or two or more units may be integrated into one unit. .

When the above function is implemented in the form of a software functional unit and sold or used as a stand-alone product, it may be stored in a single processor-executable non-volatile computer-readable storage medium. Based on this understanding, the technical solution of the present invention, that is, the part contributing to related technology or a part of the technical solution may be implemented in the form of a software product, and the computer software product is stored in one storage medium, and one It includes a plurality of instructions used to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method of each embodiment of the present invention. The above-described storage media includes various media capable of storing program code, such as USB disks, mobile hard disks, ROM, RAM, magnetic disks, or optical disks.

The above description is only a specific embodiment of the present invention, and the scope of protection of the present invention is not limited thereto. Anyone skilled in the art to which this application belongs will easily understand that any changes or replacements within the technical scope disclosed in the present invention should fall within the protection scope of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the patent claims.

Embodiments of the present invention provide an image processing method and device, an electronic device, and a computer-readable storage medium, wherein a target image is acquired; determine a shielding relationship between adjacent target objects in the target image, where modal masks of adjacent target objects are connected to each other; For a target object that is shielded in the target image, based on the shielding relationship, select a target shielding object that has a shielding relationship with the shielded target object from the target image; Based on the modal mask of the target shielding object and the modal mask of the shielded target object, determine outline information of a shielded portion of the shielded target object and image information of each pixel point within the outline of the shielded portion; The modal mask of the target object includes mask information in which the target object is not shielded by another target object.

Claims (20)

As an image processing method,
Obtaining a target image;
Determining a shielding relationship between adjacent target objects in the target image - modal masks of adjacent target objects are connected to each other;
For a target object that is shielded in the target image, based on the shielding relationship, selecting a target shielding object that has a shielding relationship with the shielded target object from the target image; and
Based on the modal mask of the target shielding object and the modal mask of the shielded target object, determining outline information of a shielded portion of the shielded target object and image information of each pixel point within the outline of the shielded portion - The modal mask of the target object includes - where the target object includes mask information that is not masked by another target object,
The step of determining the shielding relationship between adjacent target objects in the target image is,
Selecting adjacent object pairs from the target image, where the adjacent object pair includes two adjacent target objects;
Complementing the outline of one target object in the adjacent object pair by another target object in the adjacent object pair, respectively; and
An image processing method comprising determining an occlusion relationship between two adjacent target objects in the pair of adjacent objects, based on the entire contour after complementation and the overall outline before complementation of each target object. delete According to paragraph 1,
An image processing method, wherein the step of complementing the outline of one target object in the adjacent object pair by another target object in the adjacent object pair is performed by a first neural network. According to paragraph 3,
The step of determining outline information of a shielded portion of the shielded target object based on the modal mask of the target shielded object and the modal mask of the shielded target object,
determining a union of modal masks of the target shielding object; and
An image comprising the step of determining, using the first neural network, contour information of a masked portion of the masked target object according to an image corresponding to the union and a modal mask of the masked target object. How to handle it. According to paragraph 3,
The first neural network is,
Obtaining a first sample image corresponding to a modal mask of a first sample object and a second sample image corresponding to a modal mask of a second sample object; and
A portion of the first sample image is masked using the second sample image, and the first sample image, the second sample image, and the masked first sample image are input to a first neural network to be trained, An image processing method, wherein the first sample image is obtained by being trained using a step of adjusting network parameters of the first neural network to be trained, targeting to complement the outline masked by the second sample image. According to clause 5,
The step of training the first neural network is,
A portion of the second sample image is masked using the first sample image, and the first sample image and the masked second sample image are input to the first neural network to be trained, so that the first sample image An image processing method further comprising adjusting network parameters of the first neural network to be trained, with the goal of keeping the outline unchanged. According to paragraph 1,
Based on the shielding relationship, the step of selecting a target shielding object that has a shielding relationship with the shielded target object from the target image,
Based on the shielding relationship, selecting at least one level of target shielding object that has a shielding relationship with the shielded target object from the target image - the target shielding object of the first level shields the shielded target object, and , the target shielding object of another level shields the target shielding object of the previous level - an image processing method comprising: In clause 7,
The step of selecting at least one level of target shielding object that has a shielding relationship with the shielded target object from the target image,
Based on the occlusion relationship, constructing a occlusion directed graph corresponding to the target image - a node in the occlusion directed graph is a target object adjacent to another target object, and an edge in the occlusion directed graph is two adjacent targets. Indicates a target object shielded from the target object - ;
Based on the shielded directed graph, determining ancestor nodes of all levels of nodes corresponding to the shielded target object; and
An image processing method comprising using a target object corresponding to an ancestor node of each determined level as a target shielding object of each level of the shielded target object. According to paragraph 1,
The step of determining image information of each pixel point in the outline of the shielded portion of the shielded target object based on the modal mask of the target shielded object and the modal mask of the shielded target object,
Based on the modal mask of the shielded target object and the outline information of the shielded portion of the shielded target object, determining a non-modal mask of the shielded target object - the non-modal mask of the target object is the target object. Contains mask information that is not shielded by another target object and mask information that the target object is shielded by another target object -;
determining a union of modal masks of the target shielding object;
determining a first intersection of the union and a non-modal mask of the masked target object;
A step of shielding a partial target image using an image corresponding to the first intersection - the partial target image is centered on the masked target object in the target image and includes a portion of pixel points around the masked target object. Contains - ; and
Based on the image corresponding to the first intersection and the partial target image after being masked, determining image information of each pixel point in the outline of the masked portion of the masked target object. Image processing method. According to clause 9,
The step of determining image information of each pixel point within the outline of the masked portion of the masked target object based on the image corresponding to the first intersection and the partial target image after being masked includes using a second neural network. An image processing method characterized in that it is executed. According to clause 10,
The second neural network is,
Obtaining a third sample image corresponding to the modal mask of the third sample object and a fourth sample image corresponding to the modal mask of the fourth sample object;
After blocking a portion of the third sample image using the fourth sample image, determining a second intersection of a modal mask of the third sample object and a modal mask of the fourth sample object; and
Based on the masked third sample image and the image corresponding to the second intersection, the target is to complement the image information of each pixel point of the masked portion of the third sample image, and the second neural network to be trained An image processing method, characterized in that it is obtained by training using steps to adjust network parameters. According to clause 11,
Based on the masked third sample image and the image corresponding to the second intersection, the target is to complement the image information of each pixel point of the masked portion of the third sample image, and the second neural network to be trained The steps to adjust network parameters are:
Obtaining a fifth sample image corresponding to the third sample object - the fifth sample image is an image centered on the third sample object, and pixels around the third sample object and the third sample object Contains points - ;
shielding a portion of the fifth sample image using an image corresponding to the second intersection; and
Based on the masked third sample image and the masked fifth sample image, the target is to complement the image information of each pixel point of the masked portion of the third sample image, and the second neural network to be trained An image processing method comprising the step of adjusting network parameters. An image processing device, comprising:
an image acquisition module configured to acquire a target image;
a shielding determination module configured to determine a shielding relationship between adjacent target objects in the target image, where modal masks of adjacent target objects are connected to each other;
an object selection module configured to select a target shielding object that has a shielding relationship with the shielded target object from the target image, based on the shielding relationship, for a shielded target object in the target image; and
A mask configured to determine, based on the modal mask of the target shielding object and the modal mask of the shielded target object, contour information of a shielded portion of the shielded target object and image information of each pixel point within the outline of the shielded portion. a processing module comprising: a modal mask of a target object comprising mask information where the target object is not masked by another target object;
The shielding determination module also:
Select adjacent object pairs from the target image - an adjacent object pair includes two adjacent target objects;
one target object in the pair of adjacent objects each complements an outline occluded by another target object in the pair of adjacent objects;
An image processing device, characterized in that it is configured to determine a shielding relationship between two adjacent target objects in the pair of adjacent objects, based on the overall contour after complementation and the overall outline before complementation of each target object. As an electronic device,
Comprising a processor, a storage medium, and a bus, wherein the storage medium stores computer-readable instructions executable by the processor, and when the electronic device is operated, the processor and the storage medium communicate through a bus, and the process is an electronic device characterized in that the image processing method according to any one of claims 1, 3 to 12 is executed by executing the computer-readable instructions. A computer-readable storage medium, comprising:
A computer program is stored in the computer-readable storage medium, and when the computer program is operated by a processor, the image processing method according to any one of claims 1, 3 to 12 is executed. A computer-readable storage medium. delete delete delete delete delete