Infrared spliced image multi-scale enhancement method based on fusion wavelet denoising

Abstract

The invention relates to the technical field of infrared image enhancement, in particular to an infrared spliced image multi-scale enhancement method based on fusion wavelet denoising. Comprising the following steps: S1, acquiring an original infrared spliced image, and constructing a Gaussian image pyramid by using a processing result of the original infrared spliced image; s2, a Laplacian image pyramid is constructed; s3, performing first-layer wavelet denoising processing on the Laplacian image pyramid to obtain the Laplacian image pyramid after the first-layer wavelet denoising processing is performed on the Laplacian image pyramid; s4, performing contrast stretching on the Laplacian image pyramid after the first layer of wavelet denoising to obtain a Laplacian image pyramid after detail enhancement; and S5, reconstructing the high-resolution image pyramid to obtain a multi-scale enhanced infrared spliced image. The image contrast can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of infrared image enhancement, and in particular to a multi-scale enhancement method for infrared spliced ​​images based on fused wavelet denoising. Background Art

[0002] Spatial positioning of a target based on infrared images and inversion of chemical composition and content information from its infrared spectra are cutting-edge technologies in infrared remote sensing, with broad application prospects in fields such as environmental remote sensing and monitoring. However, infrared images suffer from inherent shortcomings such as poor contrast, low signal-to-noise ratio, and blurred visual effects, necessitating improvements in their quality in practical applications. Existing infrared imaging spectrometers are unable to capture infrared remote sensing images covering a large field of view in a single shot, necessitating image stitching. However, the presence of stitching gaps in infrared images poses challenges for infrared image enhancement. Traditional infrared image enhancement methods include histogram equalization and improved histogram equalization. While histogram equalization enhances the contrast of the target image, it also amplifies background noise (including image stitching gap noise). Improved histogram equalization can suppress random background noise, but it cannot accurately identify stitching gap noise, making it ineffective for enhancing mosaic infrared images. Summary of the Invention

[0003] In order to solve the problem that traditional infrared image enhancement methods cannot effectively solve the low contrast and severe stitching noise of infrared stitching images, the present invention provides a multi-scale enhancement method for infrared stitching images based on fused wavelet denoising. While performing multi-scale image enhancement based on the image pyramid transform method, the image denoising is combined with wavelet transform to remove the inherent stitching noise and random dark noise of the infrared stitching image, effectively improving the image contrast.

[0004] The multi-scale enhancement method of infrared mosaic images based on fusion wavelet denoising proposed in the present invention specifically includes the following steps:

[0005] S1: Get the original infrared mosaic image, repeat the Gaussian low-pass filtering and alternate row and column downsampling process on the original infrared mosaic image l times, and obtain the Gaussian image pyramid by the following formula:

[0006]

[0007]

[0008] Among them, G l is the l-th layer image of the Gaussian image pyramid, (i, j) is the pixel position coordinate of the Gaussian image pyramid, L is the top layer number of the Gaussian image pyramid, Rl and C l are the number of rows and columns of the lth layer of the Gaussian image pyramid, ω g (m,n) Gaussian low-pass filter, where m and n are the coordinates of the pixels around the filtered pixel, and a is a constant coefficient;

[0009] S2: Combine the l-layer image of the Gaussian image pyramid and construct the Laplacian image pyramid using the following formula:

[0010]

[0011] Among them, LP l is the lth layer image of the Laplacian image pyramid, G * l+1 is the l+1th layer image of the approximate image;

[0012] S3: Using the wavelet function to perform the first-layer wavelet denoising on the Laplacian image pyramid to obtain the Laplacian image pyramid after the first-layer wavelet denoising;

[0013] S4: Use the exponential transformation method to perform contrast stretching on the Laplacian image pyramid after the first layer of wavelet denoising to obtain the Laplacian image pyramid after detail enhancement;

[0014] S5: The Laplacian image pyramid and Gaussian image pyramid after detail enhancement are combined to reconstruct a high-resolution image pyramid and obtain a multi-scale enhanced infrared stitching image.

[0015] Preferably, step S2 specifically includes the following steps:

[0016] S21: Perform interpolation and expansion processing on the lth layer image of the Gaussian image pyramid to obtain an approximate image. The lth layer image of the approximate image has the same size as the l-1th layer image of the Gaussian image pyramid:

[0017]

[0018]

[0019] Among them, G * l is an approximate image;

[0020] S22: Using the difference between the lth layer image of the Gaussian image pyramid and the l+1th layer image of the approximate image as the lth layer image of the Laplacian image pyramid, thereby completing the construction of the Laplacian image pyramid.

[0021] Preferably, step S3 specifically includes the following steps:

[0022] S31: Use wavelet function to perform multi-scale first-level wavelet decomposition on the Laplacian image pyramid to obtain low-frequency wavelet coefficients, horizontal high-frequency wavelet coefficients, vertical high-frequency wavelet coefficients and diagonal high-frequency wavelet coefficients;

[0023] S32: Suppress the vertical high-frequency wavelet coefficients, and use the suppressed vertical high-frequency wavelet coefficients, low-frequency wavelet coefficients, horizontal high-frequency wavelet coefficients and diagonal high-frequency wavelet coefficients to perform multi-scale one-dimensional wavelet reconstruction to obtain the Laplacian image pyramid after the first layer of wavelet denoising.

[0024] Preferably, the calculation formula for suppressing vertical high-frequency wavelet coefficients is:

[0025]

[0026] Wherein, cV′ is the vertical high-frequency wavelet coefficient after suppression, cV is the vertical high-frequency wavelet coefficient, and k is a constant that makes the vertical high-frequency wavelet coefficient after suppression approximately equal to the horizontal high-frequency wavelet coefficient.

[0027] Preferably, in step S4, the calculation formula for contrast stretching is:

[0028]

[0029] Among them, LP' l It is the Laplacian image pyramid after the first layer of wavelet denoising. ENHAN is a naming method and has no physical meaning.

[0030] Preferably, step S5 specifically includes the following steps:

[0031] S51: performing an addition operation on the lowest resolution image of the Gaussian image pyramid and the lowest resolution image of the detail-enhanced Laplacian image pyramid to obtain image A;

[0032] S52: performing upsampling interpolation on image A to obtain image B having the same size as the second lowest resolution image of the detail-enhanced Laplacian image pyramid;

[0033] S53: performing an addition operation on the image B and the second low-resolution image of the detail-enhanced Laplacian image pyramid to obtain image C;

[0034] S52: performing upsampling interpolation on the image C to obtain an image D having the same size as the third low-resolution image of the Laplacian image pyramid after detail enhancement;

[0035] S55: repeating steps S53-S54 until the addition operation with the highest resolution image of the detail-enhanced Laplacian image pyramid is completed, and reconstructing the high-resolution image pyramid using the images obtained by all the addition operations;

[0036] S56: The highest resolution original scale image in the top layer of the reconstructed high-resolution image pyramid is used as the multi-scale enhanced infrared stitching image.

[0037] Preferably, in order to avoid synchronously amplifying the dark noise of the infrared detector during the multi-scale enhancement of the original infrared stitched image, the images of each layer of the high-resolution image pyramid are arranged in order from low to high resolution, and soft threshold wavelet denoising is performed on each layer of the reconstructed image of the high-resolution image pyramid in sequence according to the following formula:

[0038]

[0039] T=M×[ln(N p ) / N p ] 1 / 2 (9);

[0040] Where WT(x) is the high-frequency component of the wavelet coefficients after multi-scale first-level wavelet decomposition of the Laplacian image pyramid, M is the median of the absolute values ​​of the wavelet coefficients, Np is the number of image pixels in the high-resolution image pyramid, and T is an intermediate variable for soft threshold wavelet denoising, which has no physical meaning.

[0041] Preferably, the wavelet function is a symlets wavelet function.

[0042] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0043] The present invention proposes a multi-scale enhancement method for infrared stitching images based on fused wavelet denoising. While performing multi-scale image enhancement based on the image pyramid transform method, the wavelet transform method is combined for image denoising to remove the inherent stitching noise and random dark noise of the infrared stitching image. While gradually enhancing the details of the infrared stitching image, the image noise is suppressed, the stitching gap noise of the enhanced infrared stitching image is reduced, the image contrast is effectively improved, and the image quality is significantly enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 1 is a flow chart of a multi-scale enhancement method for infrared mosaic images based on fused wavelet denoising provided in accordance with an embodiment of the present invention;

[0045] Figure 2 is an original infrared stitched image provided by an embodiment of the present invention;

[0046] Figure 3 2 is a schematic structural diagram of a six-layer Gaussian image pyramid provided according to an embodiment of the present invention;

[0047] Figure 4 4 is a schematic structural diagram of a five-layer Laplacian image pyramid provided according to an embodiment of the present invention.

[0048] Figure 5 is the first layer image of the five-layer Laplacian image pyramid provided by an embodiment of the present invention;

[0049] Figure 6 is the first-level wavelet coefficient of the first-layer image of the five-layer Laplacian image pyramid provided by an embodiment of the present invention;

[0050] Figure 7 2 is a schematic structural diagram of a five-layer Laplacian image pyramid after detail enhancement according to an embodiment of the present invention;

[0051] Figure 8 is a schematic structural diagram of a reconstructed high-resolution image pyramid provided by an embodiment of the present invention;

[0052] Figure 9 This is a multi-scale enhanced infrared stitching image provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, identical modules are denoted by identical reference numerals. In the case of identical reference numerals, their names and functions are also identical. Therefore, their detailed description will not be repeated.

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0055] Aiming at the problem that infrared stitching images have low contrast and severe stitching noise, the present invention proposes a multi-scale enhancement method for infrared stitching images based on fusion wavelet denoising. The method mainly processes infrared stitching images based on image enhancement method of image pyramid transform and wavelet denoising method, which can greatly improve image contrast while significantly reducing image noise.

[0056] The multi-scale enhancement method of infrared mosaic images based on fusion wavelet denoising proposed in the present invention specifically includes the following steps:

[0057] S1: Get the original infrared mosaic image, repeat the Gaussian low-pass filtering and alternate row and column downsampling process on the original infrared mosaic image l times, and obtain the Gaussian image pyramid by the following formula:

[0058]

[0059]

[0060] Among them, G l is the l-th layer image of the Gaussian image pyramid, (i, j) is the pixel position coordinate of the Gaussian image pyramid, L is the top layer number of the Gaussian image pyramid, R l and C l are the number of rows and columns of the lth layer of the Gaussian image pyramid, ω g (m,n) Gaussian low-pass filter, where m and n are the position coordinates of the surrounding pixels centered on the filtered pixel, and a is a constant coefficient.

[0061] S2: Combine the l-layer image of the Gaussian image pyramid and construct the Laplacian image pyramid using the following formula:

[0062]

[0063] Among them, LP l is the lth layer image of the Laplacian image pyramid, G * l+1 is the l+1th layer image of the approximate image.

[0064] Step S2 specifically includes the following steps:

[0065] S21: Perform interpolation and expansion processing on the lth layer image of the Gaussian image pyramid to obtain an approximate image. The lth layer image of the approximate image has the same size as the l-1th layer image of the Gaussian image pyramid:

[0066]

[0067]

[0068] Among them, G * l is an approximate image;

[0069] S22: Using the difference between the lth layer image of the Gaussian image pyramid and the l+1th layer image of the approximate image as the lth layer image of the Laplacian image pyramid, thereby completing the construction of the Laplacian image pyramid.

[0070] S3: Using the wavelet function to perform the first-layer wavelet denoising on the Laplacian image pyramid to obtain the Laplacian image pyramid after the first-layer wavelet denoising;

[0071] Step S3 specifically includes the following steps:

[0072] S31: Use wavelet function to perform multi-scale first-level wavelet decomposition on the Laplacian image pyramid to obtain low-frequency wavelet coefficients, horizontal high-frequency wavelet coefficients, vertical high-frequency wavelet coefficients and diagonal high-frequency wavelet coefficients;

[0073] S32: Suppress the vertical high-frequency wavelet coefficients, and use the suppressed vertical high-frequency wavelet coefficients, low-frequency wavelet coefficients, horizontal high-frequency wavelet coefficients and diagonal high-frequency wavelet coefficients to perform multi-scale one-dimensional wavelet reconstruction to obtain the Laplacian image pyramid after the first layer of wavelet denoising.

[0074] The calculation formula for suppressing vertical high-frequency wavelet coefficients is:

[0075]

[0076] Wherein, cV′ is the vertical high-frequency wavelet coefficient after suppression, cV is the vertical high-frequency wavelet coefficient, and k is a constant that makes the vertical high-frequency wavelet coefficient after suppression approximately equal to the horizontal high-frequency wavelet coefficient.

[0077] The wavelet function is the symlets wavelet function.

[0078] S4: The exponential transformation method is used to perform contrast stretching on the Laplacian image pyramid after the first layer of wavelet denoising to obtain the Laplacian image pyramid after detail enhancement.

[0079] In step S4, the calculation formula for contrast stretching is:

[0080]

[0081] Among them, LP' l It is the Laplacian image pyramid after the first layer of wavelet denoising. ENHAN is a naming method and has no physical meaning.

[0082] S5: The Laplacian image pyramid and Gaussian image pyramid after detail enhancement are combined to reconstruct a high-resolution image pyramid and obtain a multi-scale enhanced infrared stitching image.

[0083] Step S5 specifically includes the following steps:

[0084] S51: performing an addition operation on the lowest resolution image of the Gaussian image pyramid and the lowest resolution image of the detail-enhanced Laplacian image pyramid to obtain image A;

[0085] S52: performing upsampling interpolation on image A to obtain image B having the same size as the second lowest resolution image of the detail-enhanced Laplacian image pyramid;

[0086] S53: performing an addition operation on the image B and the second low-resolution image of the detail-enhanced Laplacian image pyramid to obtain image C;

[0087] S52: performing upsampling interpolation on the image C to obtain an image D having the same size as the third low-resolution image of the Laplacian image pyramid after detail enhancement;

[0088] S55: repeating steps S53-S54 until the addition operation with the highest resolution image of the detail-enhanced Laplacian image pyramid is completed, and reconstructing the high-resolution image pyramid using the images obtained by all the addition operations;

[0089] S56: The highest resolution original scale image in the top layer of the reconstructed high-resolution image pyramid is used as the multi-scale enhanced infrared stitching image.

[0090] In order to avoid synchronously amplifying the dark noise of the infrared detector during the multi-scale enhancement of the original infrared mosaic image, the images of each layer of the high-resolution image pyramid are arranged in order from low to high resolution, and the soft threshold wavelet denoising is performed on each layer of the reconstructed image of the high-resolution image pyramid in turn according to the following formula:

[0091]

[0092] T=M×[ln(N p ) / N p ] 1 / 2 (9);

[0093] Where WT(x) is the high-frequency component of the wavelet coefficients after multi-scale first-level wavelet decomposition of the Laplacian image pyramid, M is the median of the absolute values ​​of the wavelet coefficients, Np is the number of image pixels in the high-resolution image pyramid, and T is an intermediate variable for soft threshold wavelet denoising, which has no physical meaning.

[0094] Taking the establishment of a six-layer Gaussian image pyramid as an example, the infrared mosaic image multi-scale enhancement method based on fusion wavelet denoising proposed in the embodiment of the present invention is used to Figure 2 The original infrared stitched image shown in the figure is processed to obtain the following Figure 3The six-layer Gaussian image pyramid shown in FIG is used to interpolate and expand the image layer l of the six-layer Gaussian image pyramid to obtain an approximate image, and then the six-layer Gaussian image pyramid and the approximate image are used to construct the following Figure 4 The five-layer Laplacian image pyramid shown in FIG is processed by wavelet denoising on the first layer of the five-layer Laplacian image pyramid using the wavelet function to obtain the five-layer Laplacian image pyramid after the first layer of wavelet denoising. The first layer image and the first-level wavelet coefficients of the five-layer Laplacian image pyramid are respectively as follows: Figure 5 and Figure 6 As shown in the figure, the exponential transformation method is used to perform contrast stretching on the five-layer Laplacian image pyramid after the first layer of wavelet denoising, and the following is obtained: Figure 7 The five-layer Laplacian image pyramid after detail enhancement shown in FIG is then combined with the five-layer Laplacian image pyramid after detail enhancement and the six-layer Gaussian image pyramid to reconstruct the high-resolution image pyramid. Figure 8 The highest resolution original scale image in the top layer of the reconstructed high-resolution image pyramid is used as the multi-scale enhanced infrared stitching image, and the multi-scale enhanced infrared stitching image is as shown in Figure 9 shown.

[0095] will be as Figure 9 The multi-scale enhanced infrared mosaic image shown is similar to the Figure 2 The original infrared stitched image is compared with the original infrared stitched image shown in Table 1. The comparison results are shown in Table 1. It can be seen from Table 1 that the multi-scale enhancement method for infrared stitched images based on fusion wavelet denoising provided by the embodiment of the present invention can effectively enhance the image contrast and eliminate the stitching gap noise of the original infrared stitched image.

[0096] Table 1

[0097]

[0098] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved. This is not limited herein.

[0099] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A multi-scale enhancement method for infrared mosaic images based on fusion wavelet denoising, characterized in that: The specific steps include: S1: Obtain the original infrared mosaic image, repeatedly perform Gaussian low-pass filtering and alternate row and column downsampling on the original infrared mosaic image l times, and obtain the Gaussian image pyramid by the following formula: Among them, G l is the l-th layer image of the Gaussian image pyramid, (i, j) is the pixel position coordinate of the Gaussian image pyramid, L is the top layer number of the Gaussian image pyramid, R l and C l are the number of rows and columns of the lth layer of the Gaussian image pyramid, ω g (m,n) Gaussian low-pass filter, where m and n are the coordinates of the pixels around the filtered pixel, and a is a constant coefficient; S2: Combine the image layer l of the Gaussian image pyramid and construct a Laplacian image pyramid using the following formula: Among them, LP l is the lth layer image of the Laplacian image pyramid, G * l+1 is the l+1th layer image of the approximate image; S3: performing a first-layer wavelet denoising process on the Laplacian image pyramid using a wavelet function to obtain a Laplacian image pyramid after the first-layer wavelet denoising; S4: Use the exponential transformation method to perform contrast stretching on the Laplacian image pyramid after the first layer of wavelet denoising to obtain the Laplacian image pyramid after detail enhancement; S5: Reconstructing a high-resolution image pyramid by combining the detail-enhanced Laplacian image pyramid and the Gaussian image pyramid to obtain a multi-scale enhanced infrared stitching image.

2. The multi-scale enhancement method for infrared mosaic images based on fusion wavelet denoising according to claim 1 is characterized in that: The step S2 specifically includes the following steps: S21: Performing interpolation and expansion processing on the image at the first layer of the Gaussian image pyramid to obtain an approximate image, wherein the image at the first layer of the approximate image has the same size as the image at the first-1 layer of the Gaussian image pyramid: Among them, G * l is an approximate image; S22: Using the difference between the lth layer image of the Gaussian image pyramid and the l+1th layer image of the approximate image as the lth layer image of the Laplacian image pyramid, thereby completing the construction of the Laplacian image pyramid.

3. The multi-scale enhancement method for infrared mosaic images based on fusion wavelet denoising according to claim 1 is characterized in that: The step S3 specifically includes the following steps: S31: performing multi-scale first-level wavelet decomposition on the Laplacian image pyramid using a wavelet function to obtain low-frequency wavelet coefficients, horizontal high-frequency wavelet coefficients, vertical high-frequency wavelet coefficients, and diagonal high-frequency wavelet coefficients; S32: Suppressing the vertical high-frequency wavelet coefficients, and performing multi-scale one-dimensional wavelet reconstruction using the suppressed vertical high-frequency wavelet coefficients, the low-frequency wavelet coefficients, the horizontal high-frequency wavelet coefficients, and the diagonal high-frequency wavelet coefficients to obtain a Laplacian image pyramid after first-layer wavelet denoising.

4. The multi-scale enhancement method for infrared mosaic images based on fusion wavelet denoising according to claim 3 is characterized in that: The calculation formula for suppressing the vertical high-frequency wavelet coefficient is: Wherein, cV′ is the vertical high-frequency wavelet coefficient after suppression, cV is the vertical high-frequency wavelet coefficient, and k is a constant that makes the vertical high-frequency wavelet coefficient after suppression approximately equal to the horizontal high-frequency wavelet coefficient.

5. The multi-scale enhancement method for infrared mosaic images based on fusion wavelet denoising according to claim 1 is characterized in that: In step S4, the calculation formula for the contrast stretch is: Among them, LP' l It is the Laplacian image pyramid after the first layer of wavelet denoising. ENHAN is a naming method and has no physical meaning.

6. The multi-scale enhancement method for infrared mosaic images based on fusion wavelet denoising according to claim 1 is characterized in that: The step S5 specifically includes the following steps: S51: performing an addition operation on the lowest resolution image of the Gaussian image pyramid and the lowest resolution image of the detail-enhanced Laplacian image pyramid to obtain image A; S52: performing upsampling interpolation on the image A to obtain an image B having the same size as the second low-resolution image of the detail-enhanced Laplacian image pyramid; S53: performing an addition operation on the image B and the second low-resolution image of the detail-enhanced Laplacian image pyramid to obtain image C; S52: performing upsampling and interpolation on the image C to obtain an image D having the same size as the third low-resolution image of the detail-enhanced Laplacian image pyramid; S55: repeating steps S53-S54 until the addition operation with the highest resolution image of the detail-enhanced Laplacian image pyramid is completed, and reconstructing the high-resolution image pyramid using images obtained by all the addition operations; S56: The highest resolution original scale image in the top layer of the reconstructed high-resolution image pyramid is used as the multi-scale enhanced infrared stitching image.

7. The multi-scale enhancement method for infrared mosaic images based on fused wavelet denoising according to claim 1, characterized in that: In order to avoid synchronously amplifying the dark noise of the infrared detector during the multi-scale enhancement of the original infrared stitched image, the images of each layer of the high-resolution image pyramid are arranged in order from low to high resolution, and soft threshold wavelet denoising is performed on each layer of the reconstructed image of the high-resolution image pyramid in sequence according to the following formula: T=M×[ln(N p ) / N p ] 1 / 2 (9); Wherein, WT(x) is the high-frequency component of the wavelet coefficients after multi-scale first-level wavelet decomposition of the Laplacian image pyramid, M is the median of the absolute values ​​of the wavelet coefficients, Np is the number of image pixels in the high-resolution image pyramid, and T is an intermediate variable for soft threshold wavelet denoising and has no physical meaning.

8. The multi-scale enhancement method for infrared mosaic images based on fused wavelet denoising according to claim 1, characterized in that: The wavelet function is a symlets wavelet function.

Images