A biologically-inspired camouflage image generation method, system and device
By mimicking the camouflage behavior of chameleons, an encoder and decoder are used to generate camouflage images, solving the problems of high cost and insufficient background diversity in camouflage image generation, and achieving the generation of high-quality camouflage images.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHWEST JIAOTONG UNIV
- Filing Date
- 2025-07-23
- Publication Date
- 2026-07-14
AI Technical Summary
In the task of camouflage image generation, existing technologies suffer from high acquisition and annotation costs, a lack of large-scale data, which limits research progress, and existing methods are unable to generate natural camouflage images with diverse backgrounds.
Using a bio-inspired approach that mimics the camouflage behavior of chameleons, the system extracts stylized feature maps at multiple scales through an encoder, calculates the average feature value of the background region to render the foreground object, and then uses a decoder to stitch them together to generate a camouflage image.
While maintaining background diversity, it significantly enhances the realism and blending effect of foreground objects, generating high-quality, highly concealed camouflage images.
Smart Images
Figure CN120877022B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of camouflage images, and in particular to a bio-inspired method, system, and device for generating camouflage images. Background Technology
[0002] Camouflage images are created by blending the camouflaged target with the visual features of its surroundings, such as color and texture, resulting in low visual difference between the target and the background, making it difficult to identify. They have been applied in various fields, such as autonomous driving, fruit ripeness detection, and military reconnaissance. However, the high cost of data collection and annotation, coupled with a lack of large-scale data, has limited research progress in this field.
[0003] In camouflage image generation tasks, methods can be categorized into two types: those based on background rendering of the foreground and those based on foreground generation of the background. The first method pastes the object onto another background, gradually hiding it by altering its features. However, this pasting operation disrupts spatial and semantic continuity between images, resulting in unnatural-looking images. Another approach involves building an external knowledge base and retrieving relevant background features from foreground features to generate the background. However, when suitable background features cannot be found in the external knowledge base, camouflage image generation fails. Furthermore, due to the limited information contained in foreground features, the generated backgrounds are often too simplistic and lack diversity. Summary of the Invention
[0004] The purpose of this application is to provide a bio-inspired camouflage image generation method, system, and device. By mimicking the camouflage behavior of chameleons in nature and incorporating bio-inspired design concepts, it can significantly enhance the realism and blending effect of foreground objects while maintaining background diversity, thereby generating high-quality and highly concealed camouflage images. This application can also improve the computing efficiency of edge computing devices and servers.
[0005] To achieve the above objectives, this application provides the following solution:
[0006] In a first aspect, this application provides a bio-inspired method for generating camouflage images, including:
[0007] Obtain the image to be disguised;
[0008] The image to be disguised is input into an encoder to obtain stylized feature maps at multiple scales. The encoder is used to extract features from the image to be disguised to obtain feature maps at multiple scales. The average feature value of the background region in the feature map at each scale is calculated, and the foreground object in the feature map at each scale is rendered based on the average feature value to obtain stylized feature maps at multiple scales. The feature map includes the background region and the foreground object.
[0009] Stylized feature maps at multiple scales are input into the decoder and stitched together to obtain the camouflaged image.
[0010] Secondly, this application provides a bio-inspired camouflage image generation system, comprising:
[0011] The module for acquiring the image to be disguised is used to acquire the image to be disguised.
[0012] A stylized feature map acquisition module is used to input the image to be disguised into an encoder to obtain stylized feature maps at multiple scales; the encoder is used to extract features from the image to be disguised to obtain feature maps at multiple scales, calculate the average feature value of the background region in the feature map at each scale, and render the foreground object in the feature map at each scale based on the average feature value to obtain stylized feature maps at multiple scales; the feature map includes the background region and the foreground object.
[0013] The camouflage image acquisition module is used to input stylized feature maps of multiple scales into the decoder for stitching to obtain a camouflage image.
[0014] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described bio-inspired camouflage image generation method.
[0015] According to the specific embodiments provided in this application, this application has the following technical effects:
[0016] This application provides a bio-inspired camouflage image generation method, system, and device. It extracts low-level features such as color and edges, as well as high-level features such as complex texture and semantics from the image to be camouflaged using feature maps at multiple scales. The foreground object is rendered using the average feature value of the background region in the feature maps. While preserving the background region, the color and texture information of the background region is transferred to the foreground object to generate a camouflage image that seamlessly blends with the background. This invention can significantly improve the realism and blending effect of the foreground object while maintaining background diversity, thereby generating high-quality, highly concealed camouflage images. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1This is an application environment diagram of a bio-inspired camouflage image generation method according to an embodiment of this application;
[0019] Figure 2 A flowchart illustrating a bio-inspired camouflage image generation method provided in an embodiment of this application;
[0020] Figure 3 A structural diagram of a camouflage image generator in a bio-inspired camouflage image generation method provided in another embodiment of this application;
[0021] Figure 4 A structural diagram of a fine-grained style rendering module in a bio-inspired camouflage image generation method provided in another embodiment of this application;
[0022] Figure 5 A structural diagram of a camouflage image discriminator in a bio-inspired camouflage image generation method provided in another embodiment of this application;
[0023] Figure 6 A schematic diagram of the functional modules of a bio-inspired camouflage image generation method provided in another embodiment of this application;
[0024] Figure 7 This is a schematic diagram of the camouflaged image result obtained in this application;
[0025] Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0027] To make the objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0028] The bio-inspired camouflage image generation method provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send the image to be disguised to server 104. After receiving the image, server 104 inputs it into an encoder to obtain stylized feature maps at multiple scales; the stylized feature maps at multiple scales are then input into a decoder for stitching to obtain the disguised image. Server 104 can then feed back the obtained disguised image to terminal 102.
[0029] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.
[0030] In one exemplary embodiment, such as Figure 2 As shown, a bio-inspired camouflage image generation method is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 203. Wherein:
[0031] Step 201: Obtain the image to be disguised.
[0032] Step 202: Input the image to be disguised into the encoder to obtain stylized feature maps at multiple scales; the encoder is used to extract features from the image to be disguised to obtain feature maps at multiple scales, calculate the average feature value of the background region in the feature map at each scale, and render the foreground object in the feature map at each scale based on the average feature value to obtain stylized feature maps at multiple scales; the feature map includes the background region and the foreground object.
[0033] Step 203 involves inputting stylized feature maps at multiple scales into the decoder for concatenation to obtain the camouflaged image. The encoder and decoder in this application together constitute the camouflaged image generator.
[0034] By implementing steps 201 to 203 above, this application can significantly improve the realism and blending effect of foreground objects while maintaining background diversity, thereby generating high-quality, highly concealed camouflage images.
[0035] In another exemplary embodiment of this application, such as Figure 3 As shown, after step 201, the method further includes: horizontally segmenting the image to be disguised using a horizontal feature alignment module. The image to be disguised is uniformly scaled to 256*256 pixels and horizontally divided into 16 images of size 256*16 pixels. This allows the disguised object to be disguised within the smallest and closest possible area, enhancing the camouflage effect.
[0036] In another exemplary embodiment of this application, step 202 is replaced by steps 301 to 302:
[0037] Step 301: Input the image to be disguised into the feature extraction module to obtain feature maps at multiple scales.
[0038] The feature extraction module uses the Visual Geometry Group-19 (VGG-19) model to extract features from the image to be disguised, resulting in feature maps at four scales. These four scale feature maps include two shallow feature maps (ReLU-1_1 and ReLU-2_1) and two deep feature maps (ReLU-3_1 and ReLU-4_1). The shallow feature maps are used to obtain low-level features such as color and edges of the image to be disguised. The deep feature maps are used to obtain high-level features such as texture and semantics of the image to be disguised.
[0039] This application adopts Represents feature maps at multiple scales; where, For the first Feature maps at various scales The image to be disguised. The index is the number of the scale.
[0040] Step 302: Input the feature maps at multiple scales into the fine-grained style rendering module to obtain stylized feature maps at multiple scales.
[0041] In another exemplary embodiment of this application, for feature maps of multiple scales, since only one image is used, convolution cannot be directly used to extract the complete background region, so the background feature map is determined first. For example... Figure 4 As shown, Norm represents standardization, Mean represents the mean, Std represents the variance, and MAP represents the average eigenvalue filling. Step 302 is replaced by the following steps 401-404:
[0042] Step 401: Segment the image to be disguised to obtain a binary mask of the foreground object and a binary mask of the background region of the image to be disguised; the image to be disguised includes a background region and a foreground object.
[0043] Step 402: For any scale feature map, based on the scale, downsample the binary mask of the foreground object in the image to be camouflaged to obtain the downsampled foreground object mask.
[0044] Step 403: Based on the downsampled foreground object mask and the feature map at the specified scale, determine the background feature map at the specified scale. The input sampled foreground object mask is used to mark the position of the foreground object.
[0045] Step 404: Based on the background feature map and the feature map of the scale, obtain the stylized feature map of the scale.
[0046] In another exemplary embodiment of this application, the following formula is used to determine the first... Background feature maps at various scales:
[0047] .
[0048] in, For the first Background feature maps at various scales For the first Feature maps at various scales The width dimension of the feature map. h The height dimension of the feature map. x and y For the first The position of pixels in the feature map at each scale For the first Feature maps at various scales ( x,y The eigenvalue at point ) The image to be camouflaged after downsampling Foreground object mask at a certain scale, The image to be camouflaged after downsampling Foreground objects at various scales in ( x,y The mask value at point ).
[0049] use The average feature value of the background region is obtained, where the numerator represents the sum of the feature values of all pixels in the background region, and the denominator represents the number of pixels in the background region. For the ... Feature maps at each scale For each pixel location in the background region, calculate its average feature value, and then use the calculated average feature value to fill the downsampled foreground object mask in the image to be disguised. The corresponding part is used to obtain the background feature map.
[0050] In another exemplary embodiment of this application, such as Figure 4As shown, step 404 is replaced by the following steps 501 to 503:
[0051] Step 501: Determine the mean and variance of the background feature map at the specified scale.
[0052] Step 502: Standardize the feature map at the specified scale to obtain a standardized feature map.
[0053] Step 503: Based on the standardized feature map, the mean of the background feature map at the scale, and the variance of the background feature map at the scale, obtain the stylized feature map at the scale.
[0054] After obtaining the stylized feature map at the stated scale, this application further uses a binary mask of the background region of the camouflaged image to refill the background region in the stylized feature map.
[0055] In another exemplary embodiment of this application, the following formula is used to obtain the first... Stylized feature maps at various scales: .in, For the first Stylized feature maps at various scales For the first Variance of background feature maps at each scale For the first Feature maps at various scales For the first The mean of the feature maps at each scale For the first The variance of feature maps at each scale For the first The mean of the background feature maps at each scale. The image to be camouflaged after downsampling Foreground object mask at a certain scale, This represents element-wise multiplication.
[0056] In another exemplary embodiment of this application, an encoder-decoder architecture is used for camouflage image generation, wherein the decoder hierarchy is parametrically mirrored to the encoder. Stylized feature maps of multiple scales are input to the corresponding levels of the decoder. At the entry point of each level of the decoder, stylized features of the same scale are concatenated along the channel dimension through a multi-level skip connection mechanism. At the decoder output, the block-processed results are concatenated horizontally. The camouflage image is obtained using the following formula:
[0057] .
[0058] .
[0059] .
[0060] in, Decoded features are obtained by passing stylized feature maps of multiple scales through skip connections and a decoder. The number of horizontally segmented blocks into which the image to be disguised is located. D For decoder, For splicing operations, This is a stylized feature map at scale 1. This is a stylized feature map at the second scale. This is a stylized feature map at the third scale. This is a stylized feature map at the fourth scale. This is the fully stitched feature map obtained by horizontally concatenating the decoded features. This is a convolutional kernel that does not change the size of the fully concatenated feature map, but it processes the details at the concatenation points and optimizes the overall fully concatenated feature map. This is a disguised image.
[0061] In another exemplary embodiment of this application, such as Figure 5 As shown, to improve the quality of generated camouflage images, an adversarial training method is used to identify the camouflage image discriminator. Specifically, the camouflage image is segmented into small blocks, and if a block contains a foreground object, it is identified. This training process can be represented as:
[0062] .
[0063] .
[0064] in, For the adversarial loss of the spoofing image generator, For the adversarial loss of the spoofed image discriminator, This is a camouflage image discriminator composed of a convolutional neural network. Specifically, the camouflage image generator produces camouflage images. The goal is to fool the fake image detector. The mask corresponding to the foreground object. The image to be disguised is used. The degree of disguise is improved through adversarial training between the disguised image generator and the disguised image discriminator.
[0065] Based on the same inventive concept, such as Figure 6 As shown in the illustration, this application also provides a bio-inspired camouflage image generation system. The system includes:
[0066] The image to be disguised acquisition module 601 is used to acquire the image to be disguised.
[0067] The stylized feature map acquisition module 602 is used to input the image to be disguised into the encoder to obtain stylized feature maps at multiple scales; the encoder is used to extract features from the image to be disguised to obtain feature maps at multiple scales, calculate the average feature value of the background region in the feature map at each scale, and render the foreground object in the feature map at each scale based on the average feature value to obtain stylized feature maps at multiple scales; the feature map includes the background region and the foreground object.
[0068] The camouflage image acquisition module 603 is used to input stylized feature maps of multiple scales into the decoder for stitching to obtain a camouflage image.
[0069] In another exemplary embodiment of this application, such as Figure 7 The image shown is the result of using the camouflaged image obtained in this application. Specifically, the image to be camouflaged... and its corresponding mask After horizontal segmentation, feature extraction, fine-grained style rendering, and decoding, a camouflage image is obtained. .
[0070] The bio-inspired camouflage image generation method of this application imitates the camouflage behavior of chameleons in nature, and completes the camouflage of foreground objects in an image by dynamically adjusting their color mode and other properties to match the background area.
[0071] This application horizontally segments the image to be camouflaged and aligns it along horizontal features. By dividing the image into smaller horizontal blocks, the foreground object can be camouflaged based on background features within a minimal range, enhancing the realism of the camouflage. The proposed fine-grained style rendering module preserves background features and transfers information such as color and texture from the background region to the foreground object to achieve the camouflage effect.
[0072] The horizontal feature alignment module and fine-grained style drawing module proposed in this application are flexible, plug-and-play, and can be easily adapted to other models; they are simple and efficient, without introducing any model parameters.
[0073] Compared to previous methods of background rendering foreground, this application uses only one image to generate a camouflage image, without disrupting spatial and semantic continuity, making the generated image more natural; compared to methods of foreground generating background, this application retains all background information, making the background more diverse and realistic.
[0074] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 8As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores images to be camouflaged. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a bio-inspired camouflage image generation method.
[0075] Those skilled in the art will understand that Figure 8 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0076] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0077] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0078] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A bio-inspired method for generating camouflage images, characterized in that, The method includes: Obtain the image to be disguised; The image to be disguised is input into an encoder to obtain stylized feature maps at multiple scales. The encoder is used to extract features from the image to be disguised to obtain feature maps at multiple scales. The average feature value of the background region in the feature map at each scale is calculated, and the foreground object in the feature map at each scale is rendered based on the average feature value to obtain stylized feature maps at multiple scales. The feature map includes the background region and the foreground object. Stylized feature maps at multiple scales are input into the decoder and stitched together to obtain the camouflaged image; The image to be disguised is input into the encoder to obtain stylized feature maps at multiple scales, specifically including: The image to be disguised is input into the feature extraction module to obtain feature maps at multiple scales; The feature maps at multiple scales are input into the fine-grained style rendering module to obtain stylized feature maps at multiple scales. The feature maps at multiple scales are input into the fine-grained style rendering module to obtain stylized feature maps at multiple scales, specifically including: The image to be disguised is segmented to obtain a binary mask of the foreground object in the image to be disguised; For a feature map of any scale, the binary mask of the foreground object in the image to be disguised is downsampled based on the scale to obtain the downsampled foreground object mask. Based on the downsampled foreground object mask and the feature map at the specified scale, the background feature map at the specified scale is determined; Based on the background feature map and the feature map at the specified scale, a stylized feature map at the specified scale is obtained; The camouflaged image is obtained using the following formula: in, Decoded features are obtained by passing stylized feature maps of multiple scales through skip connections and a decoder. The number of horizontally segmented blocks into which the image to be disguised is located. For decoder, For splicing operations, This is a stylized feature map at scale 1. This is a stylized feature map at the second scale. This is a stylized feature map at the third scale. This is a stylized feature map at the fourth scale. This is the fully stitched feature map obtained by horizontally concatenating the decoded features. For convolution kernel, This is a disguised image.
2. The bio-inspired camouflage image generation method according to claim 1, characterized in that, The following formula is used to determine the first Background feature maps at various scales: ； in, For the first Background feature maps at various scales For the first Feature maps at various scales The width dimension of the feature map. The height dimension of the feature map. and For the first The position of pixels in the feature map at each scale For the first Feature maps at each scale The eigenvalue at point, The image to be camouflaged after downsampling Foreground object mask at a certain scale, The image to be camouflaged after downsampling Foreground objects at various scales in ( x,y The mask value at point ).
3. The bio-inspired camouflage image generation method according to claim 1, characterized in that, Based on the background feature map and the feature map at the specified scale, a stylized feature map at the specified scale is obtained, specifically including: Determine the mean and variance of the background feature map at the specified scale; The feature map at the specified scale is standardized to obtain a standardized feature map. Based on the standardized feature map, the mean of the background feature map at the specified scale, and the variance of the background feature map at the specified scale, a stylized feature map at the specified scale is obtained.
4. The bio-inspired camouflage image generation method according to claim 3, characterized in that, The following formula is used to obtain the first... Stylized feature maps at various scales: ； in, For the first Stylized feature maps at various scales, For the first The variance of background feature maps at each scale For the first Feature maps at various scales For the first The mean of the feature maps at each scale For the first The variance of feature maps at each scale For the first The mean of the background feature maps at each scale. The image to be camouflaged after downsampling Foreground object mask at a certain scale, This represents element-wise multiplication.
5. The bio-inspired camouflage image generation method according to claim 1, characterized in that, After obtaining the image to be disguised, the method further includes: using a horizontal feature alignment module to horizontally segment the image to be disguised.
6. A bio-inspired camouflage image generation system, employing the bio-inspired camouflage image generation method according to any one of claims 1-5, the system comprising: The module for acquiring the image to be disguised is used to acquire the image to be disguised. The stylized feature map acquisition module is used to input the image to be disguised into the encoder to obtain stylized feature maps at multiple scales; the encoder is used to extract features from the image to be disguised to obtain feature maps at multiple scales, calculate the average feature value of the background region in the feature map at each scale, and render the foreground object in the feature map at each scale based on the average feature value to obtain stylized feature maps at multiple scales. The feature map includes a background region and a foreground object; The camouflage image acquisition module is used to input stylized feature maps of multiple scales into the decoder for stitching to obtain a camouflage image.
7. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the bio-inspired camouflage image generation method according to any one of claims 1-5.