Remote sensing image cloud and mist removing method based on non-sharp face mask

Abstract

The invention discloses a remote sensing image cloud and mist removing method based on a non-sharp face mask. Atmospheric light is determined by using a segmentation method so as to quantify contrast and color gray scale; a transmission image is estimated by using priori knowledge of a dark channel, the proximity between objects is determined, and finally a projected image is optimized by using a guided filtering method based on a non-sharp surface mask. An existing remote sensing image cloud removal algorithm usually has the problems of halo and artifacts. According to the remote sensing image cloud removal method based on the non-sharp face mask, a dark channel method is improved. According to the method, halo and artifacts do not need to be effectively removed, and more excellent results are obtained on various performance indexes.

Description

Technical Field

[0001] The present invention relates to a preprocessing method in remote sensing image processing, and specifically to a cloud and fog removal technology based on remote sensing images. Background Art

[0002] Remote sensing images have a wide range of applications in fields such as meteorological prediction, resource exploration, and military reconnaissance. However, limited by harsh meteorological conditions, light is often deflected and absorbed by air particles, generating cloud and rain streaks and degraded images, which often greatly reduces the visibility and contrast of remote sensing images, making the process of capturing information on the Earth's surface difficult. Therefore, atmospheric correction and cloud and fog removal are an important aspect of remote sensing image processing.

[0003] To effectively remove clouds and fog in remote sensing images, many researchers have proposed image de-clouding methods. First, there are methods based on physical models, such as the dark channel method. However, such methods have an underdetermined problem in the inversion operation due to the need to consider atmospheric light intensity and cannot produce the best results. In addition to methods based on physical models, there are also methods based on image enhancement and methods based on deep neural networks. However, the former is usually only applicable to clouds and fog of specific intensities and causes problems such as halos and artifacts, while the latter has strong data dependence and is still difficult to be applied in practice.

[0004] The present invention designs a cloud and fog removal method for remote sensing images based on an unsharp mask. Based on segmentation and an unsharp mask, a segmentation method is used to determine atmospheric light to quantify contrast and color grayscale; the prior knowledge of the dark channel is used to estimate the transmission map to determine the proximity between objects, and finally, a guided filtering method based on an unsharp mask is used to optimize the transmission map. Different from existing methods, this method improves the dark channel method and removes halos and artifacts. Summary of the Invention

[0005] To overcome the problems encountered in de-clouding remote sensing images, this paper designs a cloud and fog removal method for remote sensing images based on an unsharp mask. This technology improves on the dark channel method and removes halos and artifacts, and the proposed method has obtained better test results in various performance indicators.

[0006] The technical solution adopted by the present invention is as follows:

[0007] Step 1: Construct a degradation function and perform atmospheric transmission simulation;

[0008] Step 2: Use the dark channel method to estimate atmospheric light;

[0009] Step 3: Refine the transmission process through an unsharp mask;

[0010] Step 4: Estimate the original radiance and restore the image;

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0012] (1) This technology can avoid halos and artifacts caused by the dark channel method in cloud and fog removal;

[0013] (2) This technology has achieved higher test results in various performance indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Attached Figure 1 : Schematic diagram of the algorithm flow

[0015] Attached Figure 2 : Diagram showing the algorithm effect demonstration results DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the embodiments and the drawings.

[0017] The algorithm flow of the present invention is as Figure 1 shown. The specific method for implementing the present invention includes the following steps:

[0018] Step 1: Construct a degradation function and perform atmospheric transmission simulation

[0019] First, the atmospheric transmission matrix needs to be subdivided and evaluated. The definition of the atmospheric transmission matrix can be shown in Formula 1:

[0020]

[0021] In the formula, T hre (x) is the atmospheric transmission matrix, I i (x) is the captured input image, J ri (x) is the restored image, A l is the atmospheric light, and c is the index of the corresponding band. Usually, there is a relationship shown in Formula 2:

[0022] I i (x) = J ri (x)t(x) - A l (1 - t(x)) (2)

[0023] In the formula, I i (x) is the captured input image, J ri (x) is the restored image, A l is the atmospheric light, t(x) is the atmospheric transmission matrix. And the present invention uses the guided filtering method to estimate the atmospheric transmission map, and its estimation method is shown in Formula 3:

[0024]

[0025]

[0026] In the formula, F represents the transformation matrix for guiding image transformation, and W k (F) is the filtering transformation operator, ω represents the transformation weight, represents the directly estimated transmission map, which is shown in Formula 4 as follows:

[0027]

[0028] In the formula, I i (x) is the captured input image, Ω(x) is the weight matrix, ω represents the transformation weight, and A l is the atmospheric light.

[0029] Step 2: Use the dark channel method to estimate the atmospheric light

[0030] Formula 2 can be rewritten in the form of Formula 5:

[0031]

[0032] In the formula, I i (x) is the captured input image, J ri (x) is the restored image, A l is the atmospheric light, and t(x) is the atmospheric transmission matrix. When the dark channel pixel d sd (x,y) approaches infinity, then there is a relationship as shown in Formula 6:

[0033]

[0034] In the formula, I i (x) is the captured input image, c is the index of the corresponding band, d sd is the dark channel pixel, is the estimated atmospheric light. In the present invention, it is not assumed that d sd (x,y)→∞ is always satisfied, but the estimation is performed according to the relationship as shown in Formula 7:

[0035]

[0036] In the formula represents the intensity range of the pixel signal in the scene, and is used to find the maximum pixel signal intensity within the threshold range. Therefore, there is also a relationship as shown in Formula 8:

[0037]

[0038] Step 3: Refine the transmission process through an unsharp mask

[0039] After estimating the atmospheric light, a refined analysis is performed on the transmission matrix representing the transmission process. The present invention uses the method in Equation 9 to calculate the transmission map:

[0040]

[0041] where I(x) represents the captured input image, A represents the atmospheric light, represents the directly estimated transmission map. When there is no cloud or fog in the image, the dark channel is close to 0, that is, there is a relationship as shown in Equation 10:

[0042]

[0043] where J dark (x) represents the restored image of the dark channel.

[0044] After the initial calculation of the transmission matrix is completed, the details of the image after a finite number of transformations are rich, but the edges are still difficult to identify. And there are still halos and artifacts on the boundary after the projection map uses the soft matting method to remove artifacts. Therefore, the present invention applies a method based on an unsharp mask for guided filtering to achieve the best effect, and its process is as shown in Equation 11:

[0045]

[0046] where, T ref (F) is the refined transmission matrix, ω is a parameter, a k is a constant, G represents the image edge, represents the sharpened edge, represents the guidance window. This method can determine that the guided filter completes edge preservation and model transfer, and the guidance structure is also brought into the filtered output image to optimize the image edge.

[0047] Step 4: Estimate the original radiance and restore the image

[0048] After analyzing and estimating the atmospheric light and the transmission map of the transmission process, the original radiance in the scene can be calculated, as shown in Equation 12:

[0049]

[0050] where J c (x, y) represents the original radiance of the image, I is the captured input image, A is the atmospheric light, c is the index of the corresponding band, T ref (F) is the refined transmission matrix, t o is a constant greater than 0.1. Thus, the original radiance of the image can be restored, the atmospheric correction of the image is completed, and the cloud and fog in the image are removed.

[0051] AsFigure 2 As shown, it is the cloud removal effect of the present invention under cloud images. It can be seen that the method in the present invention can better remove clouds in remote sensing images and restore the color contrast of the images.

[0052] The above are only specific embodiments of the present invention. Any feature disclosed in this specification, unless specifically described, can be replaced by other equivalent or alternative features with similar purposes; all the disclosed features, or all the steps in all the methods or processes, except for the mutually exclusive features or / and steps, can be combined in any way.

Claims

1. A remote sensing image cloud removal method based on non-sharp mask, characterized in that: The following steps are involved: Step 1: Construct a degradation function and perform atmospheric transmission simulation; Step 2: This step is the core content of the patent; Use the dark channel method to estimate atmospheric light; Step 3: This step is the core content of the patent; the transmission process is refined through a non-sharp mask; Step 4: Estimate the original radiance and restore the image.

2. The method according to claim 1, characterized in that: In step 2, the error d sd The assumption is made that (x,y) approaches infinity, and the atmospheric light value is determined by its relationship with the original transmission matrix.

3. The method according to claim 1, characterized in that: In step 3, a non-sharp mask is used to refine the atmospheric transport process.

Images