A fixed pattern noise removal method based on minimum entropy

Abstract

The present invention discloses a method for removing fixed pattern noise based on minimum entropy, which belongs to the field of image processing. Aiming at the shortcomings of existing image denoising methods in processing fixed pattern noise, the implementation method of the present invention is as follows: first, the fixed pattern noise is modeled as the product of a calibrated noise pattern and an unknown scale parameter, effectively processing non-Gaussian and non-stationary noise; then, the scale parameter is automatically estimated using the minimum entropy principle to ensure the quality of the estimation; finally, the noise pattern scaled by the scale parameter is subtracted from the noisy image to obtain a denoised image, thereby achieving the purpose of effectively removing fixed pattern noise in the image. The present invention is applicable to fields such as image processing, can effectively correct fixed pattern noise in images, greatly improves imaging quality under dark light conditions, and has high application value.

Description

Technical Field

[0001] The invention relates to a fixed pattern noise removal method based on minimum entropy, and belongs to the field of visible light image processing. Background Art

[0002] Due to the limitation of processing technology, there is widespread non-uniformity between pixels in imaging detectors, which will lead to fixed pattern noise in the final image. The so-called fixed pattern noise can be considered as a special kind of "noise". For a given detector, the pattern of this "noise" is fixed and the intensity varies with the ambient temperature and exposure time (for a detailed introduction to fixed pattern noise, see European Machine Vision Association. Standard for characterization of image sensors and cameras [J]. EMVA Standard, 2010, 1288).

[0003] The presence of fixed pattern noise greatly limits the camera's ability to image in low light. Unlike common random noise, fixed pattern noise is basically constant within a single exposure, so fixed pattern noise accumulates at the same rate as the signal in a single exposure. Increasing the exposure time will only amplify both the signal and the noise, and will not improve the signal-to-noise ratio of the final image. In low light conditions, the radiation intensity of the scene is relatively low, so the final image will be seriously contaminated by fixed pattern noise.

[0004] Existing image denoising methods have difficulty in performing well when dealing with fixed pattern noise. This is because fixed pattern noise often appears with "overheated pixels", which cause the distribution function of fixed pattern noise to be long-tailed, which does not conform to the general Gaussian noise assumption. In addition, due to defects in the processing technology, fixed pattern noise often has low-frequency and irregular changes in space, which causes fixed pattern noise to be usually non-stationary and does not conform to the assumption of independent and identical distribution. Removing non-Gaussian and non-stationary noise is still a huge challenge so far. Summary of the invention

[0005] In view of the shortcomings of existing image denoising methods in processing fixed pattern noise, the present invention aims to provide a fixed pattern noise removal method based on minimum entropy to achieve the purpose of removing fixed pattern noise in an image.

[0006] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0007] The invention discloses a fixed pattern noise removal method based on minimum entropy. Based on the physical characteristics of detector noise, the fixed pattern noise is modeled as the product of a noise pattern and a scale parameter, and non-Gaussian and non-stationary noise is effectively processed. Then, the scale parameter is automatically estimated using the minimum entropy principle to ensure the quality of the estimation. After the noise pattern scaled by the scale parameter is subtracted from the noisy image, a denoised image is obtained, thereby achieving the purpose of effectively removing the fixed pattern noise in the image.

[0008] The present invention discloses a method for removing fixed pattern noise based on minimum entropy, comprising the following steps:

[0009] Step 1: Obtain the noise pattern F through dark field calibration.

[0010] F is an m×n matrix, where m is the image height and n is the image width. The specific calibration method is: in a dark room, cover the camera lens cap and then collect an image, and the collected image is the noise pattern F. In order to suppress random noise, multiple images are collected and then averaged to finally obtain the noise pattern F.

[0011] Step 2: Read the noisy image Y, where the image Y is an m×n matrix.

[0012] Step 3: Initialize the scale parameter α=0.

[0013] Step 4: Define auxiliary variable Z.

[0014] The auxiliary variable Z is defined as

[0015] Z=Y-α·F, (1)

[0016] Obviously Z is also an m×n matrix.

[0017] Step 5: Update α.

[0018] The α is updated as follows

[0019]

[0020] Where δ is a real number that controls the step size of the gradient descent and its value is inversely proportional to the standard deviation of the noise pattern; g(α) is the entropy of Z.

[0021] Step 6: Repeat steps 4-5 until α converges.

[0022] Let the scale parameter before step 5 be α old , the updated scale parameter is α new , the criterion for α convergence is |α new -α old |<∈, where ∈ is a small quantity.

[0023] Step 7: Subtract the fixed pattern noise to obtain the denoised image X, thereby achieving the purpose of removing the fixed pattern noise in the image.

[0024] The X is calculated by the following formula

[0025] X=Y-α·F, (3)

[0026] Obviously, X is also an m×n matrix.

[0027] Beneficial effects:

[0028] 1. The present invention discloses a fixed pattern noise removal method based on minimum entropy, which models the fixed pattern noise as the product of a calibrated noise pattern and a scale parameter, and can effectively process non-Gaussian and non-stationary noise.

[0029] 2. The present invention discloses a fixed pattern noise removal method based on minimum entropy, which uses an entropy minimization method to estimate unknown scale parameters to ensure the quality of the estimation.

[0030] 3. The present invention discloses a fixed pattern noise removal method based on minimum entropy, which can effectively remove the fixed pattern noise in the image on the basis of achieving beneficial effects 1 and 2, and greatly improve the quality of imaging under dark light conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The present invention discloses a flow chart of a method for removing fixed pattern noise based on the minimum entropy principle.

[0032] Figure 2 It is a comparison of the visual quality of the image before and after denoising using the fixed pattern noise removal method of the present invention.

[0033] Among them, (a) (c) are images before removing fixed pattern noise; (b) (d) are images after removing fixed pattern noise. DETAILED DESCRIPTION

[0034] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. The technical problems solved by the technical solution of the present invention and the beneficial effects are also described. It should be noted that the described embodiments are only intended to facilitate the understanding of the present invention and do not have any limiting effect on it.

[0035] Embodiment 1:

[0036] This embodiment discloses a fixed pattern noise removal method based on minimum entropy, which is aimed at the shortcomings of existing image denoising methods in dealing with fixed pattern noise. The hardware test conditions of this experiment are: Intel i7 6700, 8G RAM, Matlab 2016a; the camera used is a PointGrey grayscale camera, and the exposure time is set to 300ms. δ is set to 0.1.

[0037] This embodiment first models the fixed pattern noise as the product of the noise pattern and a scale parameter, and then automatically estimates the scale parameter using the minimum entropy principle. The fixed pattern noise can be effectively removed by subtracting the noise pattern scaled by the scale parameter from the noisy image. The flowchart of this embodiment is shown in FIG. Figure 1 shown.

[0038] The present embodiment discloses a method for removing fixed pattern noise based on the minimum entropy principle, comprising the following steps:

[0039] Step 1: Obtain the noise pattern F through dark field calibration.

[0040] The F is a 512×640 matrix, where 512 is the image height and 640 is the image width. The specific calibration method is: in a dark room, cover the camera lens cap and then collect an image, and the collected image is the noise pattern F. In order to suppress random noise, 100 images are collected and then averaged to finally obtain the noise pattern F.

[0041] Step 2: Read the noisy image Y.

[0042] The Y represents a noisy image, which is a 512×640 matrix.

[0043] Step 3: Initialize the scale parameter α=0.

[0044] Step 4: Define auxiliary variable Z.

[0045] The auxiliary variable Z is defined as

[0046] Z=Y-α·F, (1)

[0047] Obviously Z is also a 512×640 matrix.

[0048] Step 5: Update α.

[0049] The α is updated as follows

[0050]

[0051] Where δ is a real number that controls the step size of the gradient descent, which is 0.1 in this embodiment; g(α) is the entropy of Z. In actual use, g(α) can be approximated using the cumulative amount of each order of Z:

[0052]

[0053] in

[0054] k2=m2,k3=m3,

[0055]

[0056]

[0057]

[0058] Where Z(i,j) represents the i-th row and j-th column element of Z.

[0059] Step 6: Repeat steps 4 and 5 until α converges.

[0060] Let the scale parameter before step 5 be α old , the updated scale parameter is α new , the criterion for α convergence is |α new -α old |<10 -3 .

[0061] Step 7: Subtract the fixed pattern noise to obtain the denoised image X, thereby achieving the purpose of removing the fixed pattern noise in the image.

[0062] The X is calculated by the following formula

[0063] X=Y-α·F, (3)

[0064] Obviously, X is also a 512×640 matrix.

[0065] In order to illustrate the effectiveness of the present invention, the visual quality of the image before and after denoising using the fixed pattern noise removal method of the present invention is compared.

[0066] Figure 2 The comparison of the visual quality of the image before and after denoising using the fixed pattern noise removal method of the present invention is shown. As can be seen from the figure, the fixed pattern noise removal method of the present invention can effectively remove the fixed pattern noise in the image, greatly improve the image quality, and has a high application value.

[0067] The specific description above further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A fixed pattern noise removal method based on minimum entropy, characterized in that: The following steps are included: Step 1: Obtain the noise pattern F through dark field calibration; The implementation method of step one is: F is an m×n matrix, where m is the image height and n is the image width. The specific calibration method is: in a dark room, the camera lens cover is put on, and then an image is collected, and the collected image is the noise pattern F. In order to suppress random noise, multiple images are collected and then averaged to finally obtain the noise pattern F. Step 2: Read the noisy image Y, where the image Y is an m×n matrix; Step 3: Initialize the scale parameter α=0; Step 4: Define auxiliary variable Z; The auxiliary variable Z in step 4 is defined as Z=Y-α·F, Z is also an m×n matrix; Step 5: Update α; The implementation method of step five is: The α is updated as follows Where δ is a real number that controls the step size of the gradient descent and its value is inversely proportional to the standard deviation of the noise pattern; g(α) is the entropy of Z; Step 6: Repeat steps 4-5 until α converges; The implementation method of step six is: Let the scale parameter before step 5 be α old , the updated scale parameter is α new , the criterion for α convergence is |α new -α old |<∈, where ∈ is a small quantity; Step 7: Subtract the fixed pattern noise to obtain the denoised image X, thereby achieving the purpose of removing the fixed pattern noise in the image; The implementation method of step seven is: The X is calculated by the following formula X=Y-α·F,(3) X is also an m×n matrix.

Images