A method and device for anti-interference of dual-color image fusion based on adaptive guided filtering
By using an adaptive guided filtering method, the problems of high computational resources and weak transferability of existing image fusion algorithms are solved, achieving low-cost and efficient image fusion and target detection, which is applicable to different sensors and environments.
Patent Information
- Application Number
- CN202510700465.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-28
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2045-05-28
AI Technical Summary
Existing image fusion methods are costly in terms of computational resources and time, and their algorithmic transferability is weak, making it difficult to adapt to different sensors and environments.
An adaptive guided filtering method is adopted, which extracts feature maps through Laplacian filtering and Gaussian filtering, performs pixel comparison and guided filtering at the same coordinates, and combines scale decomposition and adaptive fusion parameters to achieve weighted fusion of low-frequency and high-frequency components of mid-wave and long-wave images.
It reduces computational resources and time costs, improves the algorithm's versatility and applicability, achieves background removal and target preservation, enhances detection accuracy, and is suitable for real-time processing.
Smart Images

Figure CN120672585B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing and target recognition technology, specifically to a method and device for anti-interference of two-color image fusion based on adaptive guided filtering. Background Technology
[0002] Infrared imaging is a widely used image type in civil and industrial fields, such as airport nighttime surveillance and nighttime water patrol operations. Because it can penetrate fog, haze, rain, snow, and other adverse weather conditions and image in the dark, it is of great value in target detection and identification in complex environments. However, in practical applications, commonly used infrared detectors typically only image specific short, medium, and long-wavelength radiation sources, and the resulting images are often affected by various noises and interferences, resulting in weak target imaging characteristics or excessive background interference, leading to a decrease in target detection accuracy. Due to the differences in target, interference, and atmospheric transmittance across different wavelengths, fusing infrared images from different wavelengths increases the entropy of target feature information, thereby enhancing the image's anti-interference capability. For example, image fusion technology can be used to combine information from multiple medium and long-wavelength infrared images to enhance image quality and target detection accuracy, improving the automation level of nighttime surveillance and detection in relevant scenarios.
[0003] Existing image fusion methods mainly include wavelet transform-based fusion methods and deep learning-based fusion methods.
[0004] For wavelet transform-based fusion methods, the computational load of wavelet transform is large, requiring high computational resources and time costs, which may limit applications with high real-time requirements.
[0005] For deep learning-based fusion methods, the trained models are often only adapted to infrared images obtained from the current sensor. To adapt to new sensors, new images need to be acquired, resulting in weak algorithm transferability. Furthermore, acquiring large-scale mid-wave and long-wave infrared data pairs for training is not easy. Summary of the Invention
[0006] In view of the shortcomings of the existing technology, the technical problem solved by the present invention is: how to reduce the correlation with the sensor and expand the scope of application when performing image fusion, while reducing computing resources and time costs.
[0007] To achieve the above objectives, in a first aspect, embodiments of this application provide a method for anti-interference through two-color image fusion based on adaptive guided filtering, the method comprising the following steps:
[0008] Acquire the mid-wave and long-wave images that need to be fused;
[0009] After feature extraction from the mid-wave and long-wave images, feature maps of the mid-wave and long-wave images are obtained; after denoising the feature maps, low-noise feature maps are obtained; after comparing pixels at the same coordinates in the low-noise feature maps, an initial attention weight map is obtained.
[0010] After applying guided filtering to the initial attention weight map, the low-frequency component fusion weights of the mid-wave and long-wave images, as well as the high-frequency component fusion weights of the mid-wave and long-wave images, are obtained.
[0011] After scale decomposition of the mid-wave and long-wave images, the low-frequency and high-frequency components of the mid-wave image and the low-frequency and high-frequency components of the long-wave image are obtained. Based on the adaptive fusion parameters of the low-frequency components, the fusion weights of the low-frequency components are weighted and fused to obtain the low-frequency fusion result. Based on the adaptive fusion parameters of the high-frequency components, the fusion weights of the high-frequency components are weighted and fused to obtain the high-frequency fusion result.
[0012] In conjunction with the first aspect, in one implementation, the process of feature extraction from the mid-wave and long-wave images includes:
[0013] After applying Laplacian filtering to the mid-wave and long-wave images respectively, the feature maps F of the mid-wave and long-wave images are obtained. n (x,y), where n represents the feature map of the medium wave image or the feature map of the long wave image, and (x,y) represents the current pixel coordinates;
[0014] F n (x,y)=I n (x,y)*L(x,y)
[0015]
[0016] L represents the Laplace operator.
[0017] In conjunction with the first aspect, in one embodiment, the low-noise feature map F n The formula for calculating ′(x,y) is:
[0018] G is a low-pass Gaussian filter with parameters r and σ.
[0019] In conjunction with the first aspect, in one implementation, the initial attention weight graph The calculation formula is:
[0020]
[0021] In conjunction with the first aspect, in one implementation, the low-frequency component fusion weights The calculation formula is: Where I n Represents a medium-wave or long-wave image; r1 is 0.075 times the shortest side of the initial attention weight map, and ò1 is 0.4;
[0022] The high-frequency component fusion weight The calculation formula is: r2 is 0.01 times the shortest side of the initial attention weight graph, and ò2 = 10 -5 .
[0023] In conjunction with the first aspect, in one implementation, the process for determining the adaptive fusion parameters of the low-frequency components includes:
[0024] Define the image size as w×n, and the maximum pixel value of the low-frequency component B1 of the mid-wave image and the low-frequency component B2 of the long-wave image is... and The logarithmic average of the brightness values of the low-frequency components of the medium-wave image and the low-frequency components of the long-wave image. The calculation formula is:
[0025]
[0026] δ is a non-zero small quantity, and the brightness gain is defined as g. n :
[0027]
[0028] Then the adaptive fusion parameters of B1 B2's adaptive fusion parameters
[0029] The formula for calculating the low-frequency fusion result is: B=|α1W1 l B1-α2W2 l B2|;
[0030] The process for determining the adaptive fusion parameters of the high-frequency components includes:
[0031] Calculate the logarithmic average of the brightness values of the high-frequency component D1 in the medium-wave image and the high-frequency component D2 in the long-wave image.
[0032]
[0033] D1's adaptive fusion parameters D2's adaptive fusion parameters
[0034] The formula for calculating the high-frequency fusion result is as follows:
[0035] In conjunction with the first aspect, in one embodiment, the process of scale decomposition of the mid-wave image and the long-wave image includes: performing Gaussian filtering on the mid-wave image and the long-wave image respectively to obtain the low-frequency component of the mid-wave image and the low-frequency component of the long-wave image; subtracting the low-frequency component of the mid-wave image from the low-frequency component of the mid-wave image to obtain the high-frequency component of the mid-wave image; and subtracting the low-frequency component of the long-wave image from the low-frequency component of the long-wave image to obtain the high-frequency component of the long-wave image.
[0036] In conjunction with the first aspect, in one embodiment, the method further includes the following step: removing the background from the low-frequency fusion result and the high-frequency fusion result to obtain a fused image.
[0037] In conjunction with the first aspect, in one embodiment, the calculation formula for the fused image A is: A = |γD - B|; where γ is an adaptive adjustment factor, and if the average value of the pixels with the lowest brightness in the high-frequency fusion result is t... D In the low-frequency fusion results, the average brightness of the top 10% to 20% of pixels is t. B ,but
[0038] Secondly, embodiments of this application provide a two-color image fusion anti-interference device based on adaptive guided filtering. The two-color image fusion anti-interference device based on adaptive guided filtering includes a processor, a memory, and a two-color image fusion anti-interference program based on adaptive guided filtering stored in the memory and executable by the processor. When the two-color image fusion anti-interference program based on adaptive guided filtering is executed by the processor, it implements the steps of the method provided in the first aspect.
[0039] Compared with the prior art, the advantages of the present invention are as follows:
[0040] (1) Compared with learning-based algorithms that require learning based on the imaging results of specific sensors, this algorithm has stronger algorithm versatility, less correlation between the algorithm and sensor characteristics, and can be flexibly applied to medium and long-wavelength imaging sensors with different imaging modes, different manufacturers, and different optoelectronic devices.
[0041] (2) Based on adaptive guided filtering, adaptive fusion weight decision based on feature intensity can be performed in different image scale spaces. This weight is calculated by the brightness gain of the high frequency component and the low frequency component of the image, which can achieve the effect of background elimination but retaining the target. Moreover, the adaptive fusion parameters utilize the statistical characteristics of the image, reducing the algorithm limitations caused by fixed parameters.
[0042] (3) The fusion result achieves a relatively ideal background removal effect, which helps human visual attention and computer vision processing, and improves detection accuracy.
[0043] (4) The present invention uses box filtering to achieve fast solution of guided filtering. The processing time on a single-channel image of a certain size is no more than 10ms, which can realize real-time processing in the engineering field.
[0044] (5) The fusion result can be used for target recognition by a general network based on infrared image target recognition, thus decoupling recognition and image fusion. This allows the fusion algorithm to be used as a preprocessing step for other algorithms and can be flexibly applied to various detection tasks, increasing the applicability of the algorithm. Attached Figure Description
[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0046] Figure 1 This is a flowchart of the dual-color image fusion anti-interference method based on adaptive guided filtering in an embodiment of the present invention;
[0047] Figure 2 These are fused images of mid-wave and long-wave images in different scenarios in embodiments of the present invention;
[0048] Figure 3 These are the merchant ship detection results using a universal model for different images in this embodiment of the invention.
[0049] Figure 4 This is a schematic diagram of the hardware structure of the adaptive guided filtering dual-color image fusion anti-interference method device involved in the embodiments of this application. Detailed Implementation
[0050] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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, 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.
[0051] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.
[0052] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.
[0053] First, the research and development process of this invention will be briefly described.
[0054] The inventors' research revealed:
[0055] Traditional wavelet-based image fusion algorithms (such as the Chinese invention patent with publication number CN105844606A) perform wavelet decomposition on the image to be fused to obtain its low-frequency and high-frequency components, then use a custom fusion rule to fuse the frequency bands, and finally perform inverse wavelet transform on each fused frequency band to obtain the final fused image. However, wavelet-based image fusion algorithms are affected by the characteristics of different images during decomposition and fusion, leading to over-fusion or under-fusion in some cases. Secondly, it is difficult to distinguish between interference and the target object in the decomposed frequency components, resulting in the fused image failing to suppress interference and thus failing to improve the target detection rate.
[0056] Fusion algorithms based on deep learning (such as the Chinese invention patent with publication number CN113033630A) employ an encoding network to encode the image to be fused, then fuse the encoded features in the fusion layer, and finally a decoder decodes the fused feature map to obtain the final fused image. To facilitate object detection, commonly used deep learning-based fusion algorithms (such as the Chinese invention patent with publication number CN113610180A) also employ a network based on object detection or object segmentation to jointly train the fusion network. This can cause the target image of the fusion network to be influenced by the detection network, thus reducing the types of images that can be processed. Secondly, the trained model is often only adapted to infrared images obtained from the current sensor. To adapt to new sensors, new images need to be acquired, resulting in weak algorithm transferability. Furthermore, acquiring large-scale mid-wave and long-wave infrared data pairs for training is not easy.
[0057] Based on this, in a first aspect, embodiments of this application provide a method for anti-interference through two-color image fusion based on adaptive guided filtering, the steps of which include:
[0058] Step A: Obtain the mid-wave and long-wave images that need to be fused. The mid-wave and long-wave images can be full-size images or images of a specified region.
[0059] Step B: After extracting features from the mid-wave and long-wave images respectively, feature maps of the mid-wave and long-wave images are obtained; after denoising the feature maps of the mid-wave and long-wave images respectively, low-noise feature maps of the mid-wave and long-wave images are obtained; after comparing the low-noise feature maps of the mid-wave and long-wave images with pixels at the same coordinates respectively, the initial attention weight maps of the mid-wave and long-wave images are obtained.
[0060] Step C: After performing adaptive parameter guided filtering on the initial attention weight maps of the mid-wave and long-wave images respectively, the low-frequency component fusion weights and high-frequency component fusion weights of the mid-wave and long-wave images are obtained.
[0061] Step D: After scale decomposition of the mid-wave and long-wave images, the low-frequency and high-frequency components of the mid-wave image and the low-frequency and high-frequency components of the long-wave image are obtained. Based on the adaptive fusion parameters of the low-frequency components, the fusion weights of the low-frequency components of the mid-wave image and the long-wave image are weighted and fused to obtain the low-frequency fusion result. Based on the adaptive fusion parameters of the high-frequency components, the fusion weights of the high-frequency components of the mid-wave image and the long-wave image are weighted and fused to obtain the high-frequency fusion result.
[0062] Step E: After removing the background from the low-frequency fusion result and the high-frequency fusion result in step D, the fused image is obtained.
[0063] In one embodiment, the process of feature extraction for the mid-wave and long-wave images in step B includes:
[0064] After applying Laplacian filtering to the mid-wave and long-wave images respectively, the feature maps F of the mid-wave and long-wave images are obtained. n (x,y), where n represents the feature map of the medium wave image or the feature map of the long wave image. In this embodiment, n = 1, 2, where 1 represents the medium wave image and 2 represents the long wave image; (x,y) represents the current pixel coordinates.
[0065] F n (x,y)=I n (x,y)*L(x,y)
[0066]
[0067] Where L represents the Laplacian operator; the Laplacian operator can effectively extract the edge and corner features of an image.
[0068] In one embodiment, the process of determining the low-noise feature maps of the mid-wave and long-wave images in step B includes: denoising the feature maps of the mid-wave and long-wave images using a Gaussian filter to obtain the low-noise feature maps F of the mid-wave and long-wave images respectively. n The formula for calculating (x,y) is:
[0069] G is a low-pass Gaussian filter with parameters r and σ.
[0070] In one embodiment, the initial attention weight graph in step B The calculation formula is:
[0071] The size of each pixel with the same coordinate in the low-noise feature maps of medium wave and long wave is compared, and the larger value is assigned 1 and the smaller value is assigned 0.
[0072] In one embodiment, the process of determining the low-frequency component fusion weights in step C includes: using the initial attention weight map of the mid-wave image and the initial attention weight map of the long-wave image as input maps, using the mid-wave image and the long-wave image in S1 as guide maps, and performing guide filtering using a guide filter with r1 and ò1 as parameters to obtain the low-frequency component fusion weights. The calculation formula is:
[0073] Where I n Representing a medium-wave or long-wave image, as mentioned earlier, n=1 represents a medium-wave image, and n=2 represents a long-wave image. r1 is set to 0.075 times the shorter side of the input image, and ò1 is set to 0.4. Guided filtering is a locally linear optimal filtering method. It can preserve the edge information of the guide image in the output image. By guiding the original image, guided filtering can maintain the consistency of the image's spatial structure. This filtering solution process can be quickly implemented using box filtering, achieving a processing speed of 10ms per frame for a 640×640 image.
[0074] Similarly, the high-frequency component fusion weights in step C... The calculation formula is:
[0075] r2 is set to 0.01 times the shorter side, ò2 = 10 -5 .
[0076] In one embodiment, the scale decomposition process for the mid-wave and long-wave images in step D includes: performing Gaussian filtering on the mid-wave and long-wave images respectively to obtain the low-frequency components of the mid-wave and long-wave images; subtracting the low-frequency components of the mid-wave image from the low-frequency components of the mid-wave image to obtain the high-frequency components of the mid-wave image; and subtracting the low-frequency components of the long-wave image from the low-frequency components of the long-wave image to obtain the high-frequency components of the long-wave image. This decomposition method is simple to operate and is a lossless and reversible operation, with lower computational complexity compared to wavelet decomposition.
[0077] In one embodiment, the process for determining the adaptive fusion parameters of the low-frequency components of the mid-wave image and the low-frequency components of the long-wave image in step D includes:
[0078] Calculate the logarithmic average of the low-frequency components of the mid-wave image and the low-frequency components of the long-wave image. If the image size is w×n, the maximum pixel value of the low-frequency component B1 of the mid-wave image and the low-frequency component B2 of the long-wave image is... and The logarithmic average of the brightness values of the low-frequency components can be expressed as: n = 1, 2;
[0079]
[0080] Where δ is a non-zero small quantity, the brightness gain is defined as g. n n = 1, 2:
[0081]
[0082] At this time, the adaptive fusion parameters of B1 B2's adaptive fusion parameters
[0083] Based on this, the formula for calculating the low-frequency fusion result in step D is: B=|α1W1 l B1-α2W2 l B2|.
[0084] Similarly, the process for determining the adaptive fusion parameters of the high-frequency components of the mid-wave image and the long-wave image in step D includes:
[0085] Calculate the logarithmic average of the brightness values of the high-frequency component D1 in the medium-wave image and the high-frequency component D2 in the long-wave image. n = 1, 2;
[0086]
[0087] At this time, the adaptive fusion parameters of D1 D2's adaptive fusion parameters
[0088] Based on this, the formula for calculating the high-frequency fusion result in step D is as follows:
[0089] In one embodiment, in step E, the formula for calculating the fused image A is: A = |γD - B|; where γ is an adaptive adjustment factor, determined by the ratio of the brightness of some pixels in the fused image of high frequency and low frequency. If the average value of the bottom 10% to 20% of the pixels in the high frequency fusion result is t... D The average brightness of the top 10%–20% of pixels in the low-frequency fusion result image is t. B ,but
[0090] The above method is illustrated below with a specific embodiment. (See [link]). Figure 1 As shown, the specific steps of this method include:
[0091] S1. Image Input: Obtain the mid-wave and long-wave images to be fused, i.e., input the long-wave infrared image and mid-wave infrared image to be fused respectively; in one embodiment, the above-mentioned mid-wave and long-wave images can be full-size images or images of a specified area (i.e., cropping the region of interest after inputting the image).
[0092] S2. Multi-scale image decomposition: After applying Gaussian filtering to the mid-wave and long-wave images respectively, the low-frequency components of the mid-wave and long-wave images are obtained. The low-frequency components of the mid-wave and long-wave images are then subtracted to obtain the high-frequency components of the mid-wave image. Similarly, the low-frequency components of the long-wave and long-wave images are subtracted to obtain the high-frequency components of the long-wave image. This decomposition method is simple to operate and is lossless and reversible, with lower computational complexity compared to wavelet decomposition.
[0093] S3. Feature Extraction: After applying Laplacian filtering to the mid-wave and long-wave images respectively, feature maps F of the mid-wave and long-wave images are obtained. n (x,y), where n represents the feature map of the medium wave image or the feature map of the long wave image. In this embodiment, n = 1, 2, where 1 represents the medium wave image and 2 represents the long wave image; (x,y) represents the current pixel coordinates.
[0094] F n (x,y)=I n (x,y)*L(x,y)
[0095]
[0096] Where L represents the Laplacian operator; the Laplacian operator can effectively extract the edge and corner features of an image.
[0097] S4. Feature Map Denoising: After applying a Gaussian filter to denoise the feature maps of the mid-wave image and the long-wave image respectively, low-noise feature maps F of the mid-wave image and the long-wave image are obtained. n The formula for calculating (x,y) is:
[0098] G is a low-pass Gaussian filter with parameters r and σ.
[0099] S5: After comparing the low-noise feature maps of the mid-wave image and the long-wave image with pixels at the same coordinates, the initial attention weight maps of the mid-wave image and the long-wave image are obtained. The calculation formula is:
[0100] The size of each pixel with the same coordinate in the low-noise feature maps of medium wave and long wave is compared, and the larger value is assigned 1 and the smaller value is assigned 0.
[0101] S6: Using the initial attention weight maps of the mid-wave image and the long-wave image as input maps, and the mid-wave and long-wave images from S1 as guiding maps, a guiding filter with parameters r1 and ò1 is used for guiding filtering to obtain the low-frequency component fusion weights. The calculation formula is:
[0102] Where I n Representing a medium-wave or long-wave image, as mentioned earlier, n=1 represents a medium-wave image, and n=2 represents a long-wave image. r1 is set to 0.075 times the shortest side of the input image (i.e., the shortest side of the initial attention weight map), and ò1 is set to 0.4. Guided filtering is a locally linear optimal filtering method. It can preserve the edge information of the guide map in the output image. By guiding the original image, guided filtering can maintain the consistency of the image's spatial structure. This filtering solution process can be quickly implemented using box filtering, achieving a processing speed of 10ms per frame for a 640×640 image.
[0103] S7: After performing adaptive parameter guided filtering on the initial attention weight maps of the mid-wave image and the long-wave image respectively, the high-frequency component fusion weights of the mid-wave image and the high-frequency component fusion weights of the long-wave image are obtained.
[0104] Similar to S6, the high-frequency component fusion weights in S7... The calculation formula is:
[0105] r2 is set to 0.01 times the shortest side, ò2 = 10 -5 .
[0106] S8: Determine the adaptive fusion parameters for the low-frequency components of the mid-wave image and the long-wave image in S2:
[0107] Calculate the logarithmic average of the low-frequency components of the mid-wave image and the low-frequency components of the long-wave image. If the image size is w×n, the maximum pixel value of the low-frequency component B1 of the mid-wave image and the low-frequency component B2 of the long-wave image is... and The logarithmic average of the brightness values of the low-frequency components can be expressed as: n = 1, 2;
[0108]
[0109] Where δ is a non-zero small quantity, the brightness gain is defined as g. n n = 1, 2:
[0110]
[0111] At this time, the adaptive fusion parameters of B1 B2's adaptive fusion parameters
[0112] Based on the adaptive fusion parameters of the low-frequency components of the mid-wave image and the long-wave image, the fusion weights of the low-frequency components of the mid-wave image and the long-wave image are weighted and fused to obtain the low-frequency fusion result: B=|α1W1 l B1-α2W2 l B2|.
[0113] S9: Determine the adaptive fusion parameters for the high-frequency components of the medium-wave image and the long-wave image in S2; similarly to S8, the process for determining the adaptive fusion parameters for the high-frequency components of the medium-wave image and the long-wave image in S9 includes:
[0114] Calculate the logarithmic average of the brightness values of the high-frequency component D1 in the medium-wave image and the high-frequency component D2 in the long-wave image. n = 1, 2;
[0115]
[0116] At this time, the adaptive fusion parameters of D1 D2's adaptive fusion parameters
[0117] Based on the adaptive fusion parameters of the high-frequency components of the mid-wave image and the long-wave image, the fusion weights of the high-frequency components of the mid-wave image and the long-wave image are weighted and fused to obtain the high-frequency fusion result:
[0118] S10: After removing the background from the low-frequency fusion result in S8 and the high-frequency fusion result in S9, a fused image A is obtained, A = |γD - B|; where γ is an adaptive adjustment factor, determined by the ratio of the brightness of some pixels in the fused image of high-frequency and low-frequency. If the average value of the bottom 15% of pixels in the high-frequency fusion result is t... D The average brightness of the top 15% of pixels in the low-frequency fusion result image is t. B ,but See Figure 2 As can be seen, the target in the fused image of this embodiment is very obvious.
[0119] S11: Training a general infrared image target detection network. In this embodiment, target recognition and detection are performed on the fused image. A YOLOv5 target recognition network is trained on a public infrared image target recognition dataset to obtain a general infrared image target recognition training model. The background-removed image is then fed into this model for inference to obtain the target detection result. See also... Figure 3 As can be seen, compared with the detection of medium-wave and long-wave images (0 ships detected), the detection accuracy of the fused image in this embodiment is significantly improved (2 ships were detected).
[0120] It can be concluded that, compared with learning-based algorithms that require learning based on the imaging results of specific sensors, this algorithm has stronger algorithmic versatility, less correlation with sensor characteristics, and can be flexibly applied to medium and long-wavelength imaging sensors of different imaging modes, manufacturers, and optoelectronic devices. In addition, based on adaptive guided filtering, the fusion weights can be made based on features in different image scale spaces, while also preserving the spatial structural consistency of the fusion weights. The fused image has stronger interpretability based on structural semantics, achieving a more ideal background removal effect, which is helpful for human visual attention and computer vision processing.
[0121] This invention employs box filtering to achieve rapid solution of guided filtering, with a processing time of no more than 10ms on single-channel images of various sizes, enabling real-time processing in engineering fields. The fusion result can be used for target recognition by a general network based on infrared image target recognition, thus decoupling recognition and image fusion and increasing the reusability of the fusion algorithm.
[0122] Thirdly, embodiments of this application provide a dual-color image fusion anti-interference device based on adaptive guided filtering. The dual-color image fusion anti-interference device based on adaptive guided filtering can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.
[0123] Reference Figure 4 , Figure 4This is a schematic diagram of the hardware structure of the two-color image fusion anti-interference device based on adaptive guided filtering involved in the embodiments of this application. In the embodiments of this application, the two-color image fusion anti-interference device based on adaptive guided filtering may include a processor, a memory, a communication interface, and a communication bus.
[0124] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.
[0125] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces. These interfaces enable interconnection of internal components within the adaptive guided filtering-based dual-color image fusion anti-interference device, and also facilitate interconnection between the device and other devices (e.g., other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.
[0126] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.
[0127] The processor can be a general-purpose processor, which can call the adaptive guided filtering-based two-color image fusion anti-interference program stored in memory and execute the adaptive guided filtering-based two-color image fusion anti-interference method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the adaptive guided filtering-based two-color image fusion anti-interference program is called can refer to the various embodiments of the adaptive guided filtering-based two-color image fusion anti-interference method of this application, and will not be repeated here.
[0128] Those skilled in the art will understand that Figure 4 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0129] Thirdly, embodiments of this application also provide a computer-readable storage medium.
[0130] The computer-readable storage medium of this application stores a two-color image fusion anti-interference program based on adaptive guided filtering, wherein when the two-color image fusion anti-interference program based on adaptive guided filtering is executed by a processor, it implements the steps of the two-color image fusion anti-interference method based on adaptive guided filtering as described above.
[0131] The method implemented when the two-color image fusion anti-interference program based on adaptive guided filtering is executed can be referred to in the various embodiments of the two-color image fusion anti-interference method based on adaptive guided filtering in this application, and will not be repeated here.
[0132] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0133] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.
[0134] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.
[0135] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.
[0136] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.
[0137] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.
[0138] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.
[0139] The above are merely specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the scope of the claims.
Claims
1. A two-color image fusion anti-interference method based on adaptive guided filtering, characterized in that, The method includes the following steps: Acquire the mid-wave and long-wave images that need to be fused; After feature extraction from the mid-wave and long-wave images, feature maps of the mid-wave and long-wave images are obtained; after denoising the feature maps, low-noise feature maps are obtained; after comparing pixels at the same coordinates in the low-noise feature maps, an initial attention weight map is obtained. After applying guided filtering to the initial attention weight map, the low-frequency component fusion weights of the mid-wave and long-wave images, as well as the high-frequency component fusion weights of the mid-wave and long-wave images, are obtained. After scale decomposition of the mid-wave and long-wave images, the low-frequency and high-frequency components of the mid-wave image and the low-frequency and high-frequency components of the long-wave image are obtained. Based on the adaptive fusion parameters of the low-frequency components, the fusion weights of the low-frequency components are weighted and fused to obtain the low-frequency fusion result. Based on the adaptive fusion parameters of the high-frequency components, the fusion weights of the high-frequency components are weighted and fused to obtain the high-frequency fusion result. The process for determining the adaptive fusion parameters of the low-frequency components includes: Define image size as Low-frequency components of mid-wave images Low-frequency components of long-wave images The maximum pixel value is and The logarithmic average of the brightness values of the low-frequency components of the mid-wave image and the low-frequency components of the long-wave image is... The calculation formula is: ; For non-zero small quantities, the brightness gain is defined as follows: : ; but Adaptive fusion parameters , Adaptive fusion parameters ; The formula for calculating the low-frequency fusion result is as follows: ; The process for determining the adaptive fusion parameters of the high-frequency components includes: Calculate the high-frequency components of a medium-wave image High-frequency components of long-wavelength images Logarithmic mean of brightness values ; ; ; Adaptive fusion parameters , Adaptive fusion parameters ; The formula for calculating the high-frequency fusion result is as follows: .
2. The anti-interference method for dual-color image fusion based on adaptive guided filtering as described in claim 1, characterized in that, The process of feature extraction for medium-wave and long-wave images includes: After applying Laplacian filtering to the mid-wave and long-wave images respectively, feature maps of the mid-wave and long-wave images are obtained. , where n represents the feature map of the medium-wave image or the feature map of the long-wave image. Indicates the current pixel coordinates; L represents the Laplace operator.
3. The anti-interference method for dual-color image fusion based on adaptive guided filtering as described in claim 2, characterized in that, The low-noise feature map The calculation formula is: ,in Based on parameters and A low-pass Gaussian filter.
4. The anti-interference method for dual-color image fusion based on adaptive guided filtering as described in claim 3, characterized in that, The initial attention weight map The calculation formula is: 。 5. The anti-interference method for dual-color image fusion based on adaptive guided filtering as described in claim 4, characterized in that, The low-frequency component fusion weight The calculation formula is: ,in Represents a medium-wave image or a long-wave image; It is 0.075 times the shortest side of the initial attention weight graph. It is 0.4; The high-frequency component fusion weight The calculation formula is: , It is 0.01 times the shortest side of the initial attention weight graph. .
6. The anti-interference method for dual-color image fusion based on adaptive guided filtering as described in any one of claims 1 to 5, characterized in that, The process of scaling the mid-wave and long-wave images includes: performing Gaussian filtering on the mid-wave and long-wave images respectively to obtain the low-frequency components of the mid-wave image and the long-wave image; subtracting the low-frequency components of the mid-wave image from the low-frequency components of the mid-wave image to obtain the high-frequency components of the mid-wave image; and subtracting the low-frequency components of the long-wave image from the low-frequency components of the long-wave image to obtain the high-frequency components of the long-wave image.
7. The anti-interference method for dual-color image fusion based on adaptive guided filtering as described in any one of claims 1 to 5, characterized in that, The method further includes the following steps: removing the background from the low-frequency fusion result and the high-frequency fusion result to obtain a fused image.
8. The anti-interference method for dual-color image fusion based on adaptive guided filtering as described in claim 7, characterized in that, The formula for calculating the fused image A is: ;in As an adaptive adjustment factor, if the average value of the bottom 10%~20% of pixels in the high-frequency fusion result is... The average brightness of the top 10% to 20% of pixels in the low-frequency fusion results is ,but .
9. A dual-color image fusion anti-interference device based on adaptive guided filtering, characterized in that, The adaptive guided filtering-based two-color image fusion anti-interference device includes a processor, a memory, and an adaptive guided filtering-based two-color image fusion anti-interference program stored in the memory and executable by the processor. When the adaptive guided filtering-based two-color image fusion anti-interference program is executed by the processor, it implements the steps of the adaptive guided filtering-based two-color image fusion anti-interference method as described in any one of claims 1 to 8.
Citation Information
Patent Citations
Wavelet transform-based image fusion method and system thereof
CN105844606A
Infrared and visible light image deep learning fusion method based on double non-local attention models
CN113033630A
Visible light image and infrared image fusion ship classification method and device based on deep learning
CN113610180A
Infrared image and visible image fusion method based on guide filtering
CN104809734A
Method and system for generating a fusion image
WO2020103601A1
