Visible and Infrared Image Fusion Method Based on Feature Extraction

Abstract

The present invention discloses a visible light and infrared image fusion method based on feature extraction. First, the infrared image is preprocessed by a hybrid filtering algorithm. Secondly, the gray histogram of the image is extended to fill the entire gray range, highlighting the infrared target and weakening the influence of the background. Then, Gaussian image pyramids are processed for different spectral images respectively, and the significant regions of the images at different scales are obtained through Laplacian pyramids. Next, the infrared region of interest is calculated, and through the dominant information comparison strategy, the significant regions of different spectra are compared to obtain the infrared image regions of interest at each scale. Finally, the infrared regions of interest at different scales are fused into the visible light image to obtain the fusion results of different spectra. The multi-spectral image fusion method of the present invention retains complete detail information and has strong robustness, solving the problem of limited spectral detection range of a single visible light sensor.

Description

Technical Field

[0001] The present invention belongs to the technical field of image fusion, and particularly relates to a fusion method for visible light and infrared spectral images with strong robustness and high fusion detail accuracy under different scenarios. Background Art

[0002] In recent years, with the popularization of multi-source image sensors, people's observation scenarios are no longer limited to the visible light range of a single band, and multi-spectral images have greatly enriched people's observation means. At present, image fusion technology has been widely applied in military, security, medical and other fields, and has played an important role in target detection, dangerous goods inspection, tissue and organ examination, etc. Especially in the military field, the technology of information fusion with multi-spectral sensors has become a key technology in modern warfare.

[0003] Abroad, image fusion technology has gradually received extensive attention since the 1990s. The Lincoln Laboratory of the Massachusetts Institute of Technology, the Air Force Research Laboratory, the Netherlands Institute for Human Factors, etc. in the United States have all conducted in-depth research in this field; well-known foreign journals such as Infrared Physics and Technology, IEEE Transaction on Image Processing, and IEEE Transaction on Signal Processing have also continuously included the latest research results in the field of multi-spectral image fusion technology.

[0004] In China, a large number of studies on image fusion technology have also been carried out in the past thirty years. Universities such as Beijing Institute of Technology and Nanjing University of Science and Technology, and research institutes such as the Institute of Remote Sensing of the Chinese Academy of Sciences have also invested in the research of image fusion systems. Domestic related journals such as "Optics & Optoelectronic Technology", "Acta Photonica Sinica", "Infrared and Laser Engineering" and other publications have also continuously published the latest research results.

[0005] The current ideas for image fusion methods can be classified into pixel level, feature level, decision level, etc. according to the objects of action. In actual engineering applications, pixel-level image fusion methods are often used. Common image fusion methods can be divided into two categories: image fusion methods based on non-multi-scale transformation and image fusion methods based on multi-scale transformation.

[0006] There are various methods for image fusion based on non - multi - scale transformation. The simplest method is to perform weighted averaging on pixels at the same position of multi - source images. This method is easy to implement. However, the fused image is prone to image stitching marks and is likely to increase image noise, and the selection of weights is a key issue. The principal component analysis (PCA) fusion method simply replaces the first principal component of the low - resolution image with the high - resolution image, so the spectral feature information in the first principal component of the low - resolution image will be lost. The image fusion methods based on non - multi - scale transformation cannot take into account the characteristics of each band in different spectral images, so the best fusion effect cannot be achieved.

[0007] The image fusion methods based on multi - scale transformation are mainly divided into image fusion methods of wavelet transform and image pyramid transform. Wavelet transform image fusion utilizes the similarity principle that wavelet multi - resolution decomposition is similar to the multi - channel decomposition of the human visual system. The multi - scale transformation characteristics are more in line with the human visual mechanism and are suitable for image fusion. The lifting wavelet transform has the advantages of strong adaptive design and irregular sampling, and the fusion visual effect is better. However, it represents only partial directional information in the source image and still causes the loss of image detail information.

[0008] Therefore, existing image fusion methods can often only simply accumulate and enhance different spectral information without making information judgments. The limitations and problems of existing technologies are that it is difficult for image fusion algorithms to extract the dominant information in different spectral detectors for image fusion. Therefore, the obtained fusion results cannot well display the advantages of different spectral detectors, resulting in the loss of image detail information. In addition, existing image fusion methods often have limited fusion effects in complex scenarios and cannot effectively distinguish between the background and target regions for image fusion in complex backgrounds, making the image fusion effect unsatisfactory. Summary of the Invention

[0009] The object of the present invention is to provide a multi - spectral image fusion method for visible light image and infrared image fusion, in which the detail information in the fused image is completely retained and it has strong robustness in different scenarios, solving the problem that the spectral detection range of a single visible light sensor is limited.

[0010] To achieve the object of the present invention, the present invention discloses a visible light and infrared image fusion method based on feature extraction, including the following steps:

[0011] Step 1, infrared image pre - processing, pre - process the infrared image using a hybrid filtering algorithm;

[0012] Step 2, infrared image enhancement, adopt a contrast enhancement algorithm to expand the gray - level histogram of the infrared image to make it fill the entire gray - level range, highlight the infrared target and weaken the background influence, and obtain a background - suppressed infrared image;

[0013] Step 3, Salient region extraction: Perform Gaussian image pyramid processing on the infrared image and the corresponding visible light image respectively, then restore the size of each layer of the pyramid, and obtain the salient regions of the infrared image and the visible light image at different scales through Laplacian pyramids;

[0014] Step 4, Infrared region of interest calculation: Compare the salient regions of different spectra through a dominant information comparison strategy to obtain the regions of interest of the infrared image at each scale;

[0015] Step 5, Multispectral image fusion: Incorporate the infrared regions of interest at different scales into the visible light image to obtain the fusion results of different spectra.

[0016] Furthermore, Step 1 is specifically as follows:

[0017] For an image window with a size of N×N, it is represented by a hybrid filtering algorithm as:

[0018]

[0019] g1 = med{f(i, j - N),... f(i, j),... f(i, j + N)}

[0020] g2 = med{f(i - N, j),.. f(i, j),... f(i + N, j)}

[0021] g3 = med{f(i + N, j - N),... f(i, j),... f(i - N, j + N)}

[0022] g4 = med{f(i - N, j - N),... f(i, j),... f(i + N, j + N)}

[0023] In the formula, K filter (x, y) represents the processing result of the hybrid filtering at the point (x, y), f(i, j) represents the gray value at the position (i, j) in the original infrared image, N is the window size, med{} is the median operation, and a set of median methods are defined to calculate the gray medians in the four directions of 0°, 45°, 90°, and 135° of the target point respectively. Taking the median represents the gray value of the target point, then g1, g2, g3, and g4 respectively represent the gray medians in the 90°, 0°, 135°, and 45° directions at the point (i, j).

[0024] Furthermore, Step 2 is specifically as follows:

[0025] Enhance the preprocessed infrared image according to the following formula to obtain a background-suppressed infrared image,

[0026]

[0027] In the above formula, f(x, y) is the pixel gray value at the infrared image coordinates (x, y), l(x, y) is the pixel gray value at the coordinates (x, y) after stretching transformation, min[f(x, y)] represents the minimum pixel gray value of the source image, max[f(x, y)] represents the maximum pixel gray value of the source image, and N represents the gray level of the stretched image, with a value range of [0, 256].

[0028] Further, step 3 is specifically as follows:

[0029] Step 3-1: Calculate the Gaussian image pyramid. Respectively, pass the enhanced infrared image and visible light image through Gaussian filtering to obtain the Gaussian blurred image, and then obtain the first-layer Gaussian pyramid image G1 through downsampling. Then perform the above steps on the first-layer Gaussian pyramid, and the entire Gaussian pyramid image sequence G can be obtained by continuously iterating i times. i , and the two-dimensional Gaussian filtering function is shown as follows:

[0030]

[0031] In the formula, the variance of Gaussian filtering is σ, x and y respectively represent the horizontal and vertical coordinates of the two-dimensional plane, and a Gaussian pyramid image is generated by using a group of Gaussian variances of σ1 < σ2 < σ3 < σ4;

[0032] Step 3-2: Extract multi-scale features. Interpolate and upsample the Gaussian pyramid image G i to obtain the Gaussian pyramid image P with the restored size i ,

[0033] Then the Laplacian pyramid can be obtained by the following formula:

[0034] L i = G i - P i

[0035] where L i represents the i-th layer of the Laplacian pyramid;

[0036] Step 3-3: Calculate the significant region image. By calculating the Laplacian image pyramids of the obtained visible light and infrared images, the significant regions of different spectral images can be obtained respectively:

[0037]

[0038] In the formula, a i is the fusion gain coefficient of the i-th layer of the Laplacian image pyramid, b i is the fusion gain coefficient of the i-th layer of the Laplacian image pyramid, a i and bi The value range of i represents the infrared Laplacian image pyramid sequence, and Lv i represents the visible light Laplacian image pyramid sequence; The Laplacian image pyramids of each layer are weighted and fused to obtain the visible light salient region image V Ros and the infrared salient region image Ir Ros .

[0039] Furthermore, step 4 is specifically as follows:

[0040]

[0041] P(x, y) = g(x, y) ⊙ Ir Ros (x, y)

[0042] In the formula, ⊙ is defined as the multiplication of corresponding elements of the matrix, and g(x, y) represents the fusion gain coefficient of the infrared salient region; V Ros (x, y) represents the gray value at the position (x, y) of the feature extraction visible light salient image, and Ir Ros (x, y) represents the gray value at the position (x, y) of the infrared salient image; When at the point (x, y), there is V Ros (x, y) - Ir Ros (x, y) is greater than or equal to the threshold T, it means that it is in the visible light salient region, and there is no need to fuse the infrared image into the visible light image. At this time, let the fusion gain coefficient g(x, y) = 0 to achieve the purpose of retaining the detailed information in the visible light image;

[0043] When V Ros (x, y) - Ir Ros (x, y) is less than the threshold T, it means that it is in the infrared salient region. At this time, let the fusion gain coefficient g(x, y) = a, where a is a constant, and the value range is [0.5, 2]. P(x, y) represents the gray value of the infrared feature image at the point (x, y), which is equal to the product of the fusion gain coefficient g(x, y) at this point and the infrared salient image Ir Ros (x, y).

[0044] Furthermore, step 5 is specifically as follows:

[0045] Fuse the obtained infrared region of interest image into the original visible light image:

[0046] F img = I img + P img

[0047] In the formula, P img represents the image of the infrared region of interest, and I img represents the visible light image, and Fimg It represents the finally obtained fused image.

[0048] Compared with the prior art, the significant progress of the present invention lies in: 1) The multi-spectral image fusion method of the present invention retains complete detail information in the fused image and has strong robustness in different scenarios, solving the problem of limited spectral detection range of a single visible light sensor; 2) Different spectral dominant information is completely extracted. The present invention can retain the dominant information in different spectral images through a comparison fusion strategy, enabling complete retention of detail information in the fused image. Existing image fusion methods often only simply accumulate and enhance different spectral information without making information judgments, resulting in loss of image detail information. The algorithm uses a comparison fusion strategy to separately retain the dominant regions in the infrared image and the visible light image, achieving the preservation of detail information in the image fusion process and improving the spatial resolution of the fused image; 3) It has strong robustness in different scenarios. The present invention uses a Laplacian of Gaussian image pyramid to perform multi-scale decomposition on different spectral images, thereby obtaining the feature information of different frequency bands of the image, and thus distinguishing background clutter from the target region of interest. Compared with traditional image fusion methods that cannot effectively distinguish the background from the target region, the method in this paper can extract the target region in complex scenarios for information fusion and has stronger robustness in different scenarios.

[0049] To more clearly illustrate the functional characteristics and structural parameters of the present invention, the following further explains in conjunction with the accompanying drawings and specific embodiments. Brief Description of the Drawings

[0050] The accompanying drawings described herein are used to provide a further understanding of the present invention and form a part of this application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0051] Figure 1 is the main flowchart of the visible light and infrared image fusion method of the present invention;

[0052] Figure 2 is the original infrared image;

[0053] Figure 3 is the preprocessed infrared image;

[0054] Figure 4 is the infrared image after feature enhancement and background suppression;

[0055] Figure 5 is the saliency region image of the visible light and infrared images;

[0056] Figure 6 is the infrared image after extraction of the region of interest;

[0057] Figure 7 It is an image after fusing visible light and infrared regions of interest. Specific implementation manner

[0058] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0059] As Figure 1 shown, Figure 1 It is the main flowchart of the visible light and infrared image fusion method of the present invention, including the following steps:

[0060] Step 1, infrared image preprocessing, preprocess the infrared image using a hybrid filtering algorithm;

[0061] Step 2, infrared image enhancement, using a contrast enhancement algorithm, expand the gray histogram of the infrared image to fill the entire gray range, highlight the infrared target and weaken the background influence, and obtain a background-suppressed infrared image;

[0062] Step 3, saliency region extraction, perform Gaussian image pyramid processing on different spectral images respectively, then restore the size of each layer of the pyramid, and obtain the saliency regions of the infrared image and the visible light image at different scales through the Laplacian pyramid respectively;

[0063] Step 4, calculation of infrared region of interest, compare the saliency regions of different spectra through a dominant information comparison strategy to obtain the regions of interest of the infrared image at each scale;

[0064] Step 5, multi-spectral image fusion, fuse the infrared regions of interest at different scales into the visible light image to obtain the fusion results of different spectra.

[0065] Embodiment

[0066] Step 1, infrared image preprocessing, since the infrared detector is prone to non-uniformity due to its own device, it will affect the imaging quality of the infrared image, as Figure 2 shown. Therefore, it is first necessary to perform hybrid filtering on the infrared image to remove the noise influence in the infrared image and improve the signal-to-noise ratio.

[0067] For an image window with a size of N×N, it is expressed by the hybrid filtering algorithm as:

[0068]

[0069] In the formula, K filter(x, y) represents the processing result of the hybrid filter at the point (x, y), f(i, j) represents the gray value at the position (i, j) in the original infrared image, N is the window size, med{} represents the median operation. Define a set of median methods to calculate the gray medians in the four directions of 0°, 45°, 90°, and 135° of the target point respectively. Taking the median represents the gray value of the target point. Then g1, g2, g3, and g4 respectively represent the gray medians in the directions of 90°, 0°, 135°, and 45° at the target point (i, j). Through image preprocessing, the infrared image after noise removal is obtained as Figure 3 shown as follows.

[0070] Step 2: Infrared image enhancement. The contrast stretching image enhancement algorithm is used to further enhance the preprocessed infrared image to achieve the purpose of suppressing the background and highlighting the infrared target.

[0071] The preprocessed infrared image is enhanced according to the following formula to obtain the background-suppressed infrared image.

[0072]

[0073] In the above formula, f(x, y) is the pixel gray value at the infrared image coordinate (x, y), l(x, y) is the pixel gray value at the coordinate (x, y) after stretching transformation, min[f(x, y)] represents the minimum pixel gray value of the source image, max[f(x, y)] represents the maximum pixel gray value of the source image, and N represents the gray level of the stretched image, with a value range of [0, 256].

[0074] Specifically, in this embodiment, Figure 4 is the enhanced infrared image.

[0075] Step 3: Salient region extraction. Gaussian image pyramids are processed for the enhanced infrared image and the visible light image respectively to obtain images at different scales of different spectral images. Then the sizes of the Gaussian pyramid images are restored, and the Laplacian image pyramid is obtained by differentiating adjacent Gaussian images, so that the high-frequency feature information at different scales can be extracted. The high-frequency feature information at different scales of different spectra is weighted and fused respectively to obtain the salient regions of the infrared image and the visible light image. The specific steps include:

[0076] Step 3-1: Gaussian image pyramid calculation:

[0077] The original image is filtered by Gaussian filter to obtain the Gaussian-blurred image, and then the first-layer Gaussian pyramid image G1 is obtained through downsampling. Then the above steps are performed on the first-layer Gaussian pyramid, and the entire Gaussian pyramid image sequence G can be obtained by iterating i times continuously. i The two-dimensional Gaussian filter function is shown as follows:

[0078]

[0079] In the formula, the variance of Gaussian filtering is σ, and x and y represent the horizontal and vertical coordinates of the two-dimensional plane respectively. A Gaussian pyramid image is generated by using a set of Gaussian variances of σ1 < σ2 < σ3 < σ4;

[0080] Step 3-2, Multi-scale feature extraction:

[0081] Interpolate and upsample the Gaussian pyramid image G i to obtain the Gaussian pyramid image P with the size restored i . Then the Laplacian pyramid can be obtained by the following formula:

[0082] L i = G i - P i (4)

[0083] where L i represents the i-th layer of the Laplacian pyramid.

[0084] Step 3-3, Calculation of the saliency region image:

[0085] By calculating the obtained Laplacian image pyramids of visible light and infrared images, the saliency regions of different spectral images can be obtained respectively:

[0086]

[0087] In the formula, a i is the fusion gain coefficient of the i-th layer Laplacian image pyramid, b i is the fusion gain coefficient of the i-th layer Laplacian image pyramid, a i and b i have a value range of [0.5, 2]; Lr i represents the infrared Laplacian image pyramid sequence, Lv i represents the visible light Laplacian image pyramid sequence; The Laplacian image pyramids of each layer are weighted and fused to obtain the visible light saliency region image V Ros and the infrared saliency region image Ir Ros .

[0088] The saliency region extraction results are as Figure 5 shown Figure 5 (a) is the original visible light image, 5(b) is the enhanced infrared image obtained after processing in Step 1 and Step 2, Figure 5 (c) and Figure 5 (d) are the saliency region extraction results of visible light and infrared images respectively.

[0089] Step 4, Calculation of infrared region of interest: Compare the salient regions of the infrared and visible light images, find the regions of interest that are significantly present in the infrared image but not in the visible light image, and the infrared feature image is defined as the image containing the infrared region of interest. The calculation method of the infrared region of interest image is as follows:

[0090]

[0091] P(x,y) = g(x,y) ⊙ Ir Ros (x,y) (7)

[0092] In the formula, ⊙ is defined as the multiplication of corresponding elements of the matrix, g(x,y) represents the fusion gain coefficient of the infrared salient region, and V Ros (x,y) extracts the gray value at the visible light salient image (x,y) by feature extraction, and Ir Ros (x,y) represents the gray value at the infrared salient image (x,y). When at the point (x,y), if there is V Ros (x,y) - Ir Ros (x,y) is greater than or equal to the threshold T, it means that it is within the visible light salient region, and there is no need to fuse the infrared image into the visible light image. At this time, let the fusion gain coefficient g(x,y) = 0 to achieve the purpose of retaining the detailed information in the visible light image.

[0093] When V Ros (x,y) - Ir Ros (x,y) is less than the threshold T, it means that it is within the infrared salient region. At this time, let the fusion gain coefficient g(x,y) = a (a is a constant, usually taking values in [0.5, 2]). In formula (6), P(x,y) represents the gray value of the infrared feature image at the point (x,y), which is equal to the product of the fusion gain coefficient g(x,y) at this point and the infrared salient map Ir Ros (x,y).

[0094] Figure 6 The infrared region of interest extraction result is shown as follows.

[0095] Step 5, The image fusion step includes:

[0096] Fuse the obtained infrared region of interest image into the original visible light image:

[0097] F img = I img + P img (8)

[0098] In the formula, P img represents the image of the infrared region of interest, I img represents the visible light image, and F img represents the finally obtained fused image.

[0099] The fusion result of the visible light and the infrared region of interest is as Figure 7 shown.

[0100] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0101] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A visible light and infrared image fusion method based on feature extraction, characterized in that It includes the following steps: Step 1, infrared image preprocessing, preprocessing the infrared image using a hybrid filtering algorithm; Step 2, infrared image enhancement, using a contrast enhancement algorithm to expand the gray histogram of the infrared image to fill the entire gray range, highlighting the infrared target and weakening the background influence to obtain a background-suppressed infrared image; Step 3, significant region extraction, performing Gaussian image pyramid processing on the infrared image and the corresponding visible light image respectively, then restoring the size of each layer of the pyramid, and obtaining the significant regions of the infrared image and the visible light image at different scales through Laplacian pyramids respectively; Step 4, infrared region of interest calculation, comparing the significant regions of different spectra through a dominant information comparison strategy to obtain the regions of interest of the infrared image at each scale; Step 5, multispectral image fusion, fusing the infrared regions of interest at different scales into the visible light image to obtain the fusion results of different spectra; The specific dominant information comparison strategy in Step 4 is: P(x, y) = g(x, y) ⊙ Ir Ros (x, y) wherein, ⊙ is defined as the multiplication of corresponding elements of matrices, and g(x, y) represents the fusion gain coefficient of the infrared significant region; V Ros (x, y) represents the gray value at the position (x, y) of the feature extraction visible light significant image, and Ir Ros (x, y) represents the gray value at the position (x, y) of the infrared significant region image; when at the point (x, y), there is V Ros (x, y) - Ir Ros (x, y) is greater than or equal to the threshold T, it means that it is in the visible light significant region, and there is no need to fuse the infrared image into the visible light image. At this time, the fusion gain coefficient g(x, y) = 0 is set to achieve the purpose of retaining the detailed information in the visible light image; When V Ros (x,y)-Ir Ros (x,y) is less than the threshold T, it indicates being in the infrared significant region. At this time, let the fusion gain coefficient g(x,y) = a, where a is a constant with a value in [0.5, 2]. P(x,y) represents the gray value of the infrared feature image at the point (x,y), which is equal to the product of the fusion gain coefficient g(x,y) at this point and the infrared significant region image Ir Ros (x,y).

2. The visible light and infrared image fusion method based on feature extraction according to claim 1, wherein The specific content of Step 1 is: For an image window with a size of N×N, the hybrid filtering algorithm is expressed as: g1 = med{f(i, j - N),... f(i, j),... f(i, j + N)} g2 = med{f(i - N, j),.. f(i, j),... f(i + N, j)} g3 = med{f(i + N, j - N),... f(i, j),... f(i - N, j + N)} g4 = med{f(i - N, j - N),... f(i, j),... f(i + N, j + N)} Where K filter (x, y) represents the processing result of the hybrid filter at the point (x, y), f(i, j) represents the gray value at the position (i, j) in the original infrared image, N is the window size, med{} is the median operation, a set of median methods are defined, and the gray medians in the four directions of 0°, 45°, 90°, and 135° of the target point are calculated respectively. Taking the median represents the gray value of the target point. Then g1, g2, g3, and g4 respectively represent the gray medians in the directions of 90°, 0°, 135°, and 45° at the target point (i, j).

3. The visible light and infrared image fusion method based on feature extraction according to claim 1, wherein The specific content of Step 2 is: Perform infrared image enhancement on the preprocessed infrared image according to the following formula to obtain a background-suppressed infrared image, In the above formula, f(x, y) is the pixel gray value at the coordinate (x, y) of the infrared image, l(x, y) is the pixel gray value at the coordinate (x, y) after stretching transformation, min[f(x, y)] represents the minimum pixel gray value of the source image, max[f(x, y)] represents the maximum pixel gray value of the source image, and N represents the number of gray levels after stretching the image, with a value range of [0, 256].

4. The visible light and infrared image fusion method based on feature extraction according to claim 1, characterized in that The specific content of Step 3 is: Step 3-1: Gaussian image pyramid calculation. Respectively, the enhanced infrared image and visible light image are filtered by Gaussian filter to obtain the Gaussian blurred image, and then the first-layer Gaussian pyramid image G1 is obtained through downsampling. Then, the above steps are performed on the first-layer Gaussian pyramid, and the entire Gaussian pyramid image sequence G can be obtained by continuously iterating i times. i , and the two-dimensional Gaussian filtering function is shown as follows: In the formula, the variance of Gaussian filtering is σ, x and y respectively represent the horizontal and vertical coordinates of the two-dimensional plane, and a Gaussian pyramid image is generated by using a set of Gaussian variances of σ1 < σ2 < σ3 < σ4; Step 3-2, multi-scale feature extraction, perform interpolation upsampling on the Gaussian pyramid image G i to obtain the Gaussian pyramid image P with the size restored i , Then the Laplacian pyramid can be obtained by the following formula: L i = G i - P i where L i represents the i-th layer of the Laplacian pyramid; Step 3-3, significant region image calculation, by calculating the Laplacian image pyramids of the obtained visible light and infrared images, the significant regions of different spectral images can be obtained respectively: Where a i is the fusion gain coefficient of the i-th layer Laplacian image pyramid, and b i is the fusion gain coefficient of the i-th layer Laplacian image pyramid. The value ranges of a i and b i are [0.5, 2]; Lr i represents the infrared Laplacian image pyramid sequence, and Lv i represents the visible light Laplacian image pyramid sequence. By weighted fusing each layer of the Laplacian image pyramid, the visible light significant region image V Ros and the infrared significant region image Ir Ros are obtained.

5. A visible light and infrared image fusion method based on feature extraction according to claim 1, characterized in that The specific content of Step 5 is: Fuse the obtained infrared region of interest image into the original visible light image: F img = I img + P img where P img represents the image of the infrared region of interest, I img represents the visible light image, and F img represents the finally obtained fused image.

Images