Low-illumination image enhancement method based on polarization de-scattering model

Abstract

The invention discloses a low-illumination image enhancement method based on a polarization de-scattering model, and the method comprises the steps: firstly, obtaining a plurality of low-illumination images with different polarization directions, and building a maximum light intensity model and a minimum light intensity model, so as to obtain a maximum light intensity image and a minimum light intensity image; then, establishing a constraint decomposition model by means of the maximum light intensity image and the minimum light intensity image to decompose the low-illumination inversion image, and calculating a decomposed component according to an image enhancement measurement index; and finally, fusing the obtained components into a scattering degradation physical model, establishing a scattering-free image inversion model, and inverting the obtained scattering-free image to obtain a final enhancement result. According to the invention, the polarization de-scattering model is applied to the field of low-illumination image enhancement, so that the low-illumination image enhancement of a complex polarization characteristic target can be realized simply and efficiently by using polarization information; man-machine interaction is not needed, and the application range of the polarization imaging technology is expanded.

Description

Technical Field

[0001] The present invention relates to a low-light image enhancement method, and in particular to a low-light image enhancement method based on a polarization descattering model. Background Art

[0002] Imaging in low-light environments has important applications in many fields, including intelligent transportation, target detection and recognition, and video surveillance. However, enhancing low-light vision is a challenging task due to the influence of the image acquisition environment. Imaging in low-light environments results in insufficient light reflected from the target. As a result, low-light images often suffer from loss of detail, low brightness, low contrast, and color distortion, which severely impair human perception. Polarization imaging, an imaging method that provides an additional dimension of information compared to traditional intensity imaging, has proven to be a promising approach for low-light image enhancement.

[0003] However, existing methods for using polarization imaging to enhance low-light vision are mostly based on data-driven deep learning methods, which require the collection of large amounts of image data for training. The entire process is complex and time-consuming, greatly limiting the on-site application of such technologies. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to propose a low-light image enhancement method based on a polarization descattering model, which can simply and efficiently utilize polarization information to achieve low-light image enhancement of targets with complex polarization characteristics without the need for human-computer interaction.

[0005] Technical solution: The low-light image enhancement method based on the polarization descattering model described in the present invention specifically includes the following steps:

[0006] (1) Acquire multiple low-light images with different polarization directions, establish a maximum light intensity model and a minimum light intensity model to obtain a maximum light intensity image and a minimum light intensity image;

[0007] (2) Using the maximum light intensity image and the minimum light intensity image, a constrained decomposition model is established to decompose the low-light inversion image, and the decomposed components are calculated based on the image enhancement metrics;

[0008] (3) The decomposed components are integrated into the scattering degradation physical model, a descattering image inversion model is established, and the descattering image is inverted to obtain the final enhancement result.

[0009] Furthermore, the multiple low-light images with different polarization directions in step (1) are polarization images I(0), I(45), I(90) and I(135) corresponding to polarization directions of 0°, 45°, 90° and 135°.

[0010] Furthermore, the process of constructing the maximum light intensity model and the minimum light intensity model in step (1) is as follows:

[0011] Calculate the Stokes vector using the polarization image:

[0012]

[0013] Where I represents the total light intensity in the low-light scene; Q is the intensity difference between the horizontal and vertical directions; U is the intensity difference between the 45° and 135° directions;

[0014] Calculate the overall degree of polarization:

[0015]

[0016] Where DoP represents the overall degree of polarization;

[0017] Establish maximum light intensity model and minimum light intensity model:

[0018]

[0019] Among them, I max is the maximum light intensity image; I min is the minimum light intensity image.

[0020] Furthermore, the implementation process of step (2) is as follows:

[0021] With the help of the maximum light intensity image I max With the minimum light intensity image I min Establish a constrained decomposition model to decompose the low-light inverted image:

[0022] R=1-I=(XI max )+(YI min )=R max +R min (4)

[0023] Among them, R is the low light inverted image; X and Y are the values corresponding to I max with I min The decomposition coefficient of , and satisfy the relationship X+Y=1; R max For I max The corresponding decomposition component is the polarization component of the maximum light intensity of the low-light inverted image; R min For I min The corresponding decomposition component is the minimum light intensity polarization component of the low-light inverted image;

[0024] The constraint decomposition model must meet the following constraints, namely:

[0025]

[0026] Where ΔI is I max with I min The difference is the low-light polarization difference image;

[0027] Decomposed component R max With R min It is necessary to use the decomposition coefficients X and Y as independent variables for calculation, and use the image enhancement metric as the objective function to obtain the optimal solution that satisfies the maximum image enhancement metric, that is:

[0028] (X,Y) optimal =argmax{EME(ΔR)} (6)

[0029] Where ΔR is R max With R min The difference between the two images is the low-light inverted polarization difference image; EME(ΔR) is the image enhancement metric of ΔR.

[0030] Furthermore, the image enhancement metric EME(ΔR) is:

[0031]

[0032] The image ΔR is divided into k1×k2 blocks numbered (k, l); i max,k,l and i min,k,l are the maximum and minimum values in each block numbered (k, l); q is a very small number that avoids being divided by 0.

[0033] Furthermore, the implementation process of establishing the descattered image inversion model in step (3) is as follows:

[0034] Introduce the scattering degradation physical model, namely:

[0035] R=D+B=L·t+A ∞ (1-t) (8)

[0036] Where R is the low-light inversion image, which is similar to the scattering degraded image; D is the direct transmitted light; B is the backscattered light; L is the descattered image; t is the medium transmittance; A ∞ is the backscattered light intensity at infinity;

[0037] According to the polarization descattering method, the backscattered light B is:

[0038]

[0039] Where ΔR is the difference between the components; P scat is the polarization degree of backscattered light;

[0040] Combining formulas (8) and (9), the descattered image inversion model is established:

[0041]

[0042] Where L is the descattered image; A ∞ is the backscattered light intensity at infinity; P scat is the polarization degree of backscattered light; ε is the correction coefficient to ensure B<R.

[0043] Furthermore, the backscattered light polarization degree P scat Obtained by global estimation method:

[0044] The maximum intensity polarization component R of the acquired low-light inversion image max The minimum intensity polarization component R of the low-light inverted image min Perform low-pass filtering, that is:

[0045]

[0046] Among them, B max With B min R max With R min The estimated backscattered light in ; LPF{·} is low-pass filtering;

[0047] According to the definition of polarization degree, the polarization degree of backscattered light is estimated globally, namely:

[0048]

[0049] in, is the globally estimated degree of polarization of the backscattered light.

[0050] Furthermore, the backscattered light intensity at infinity is an average value of the top 0.1% pixels with the largest grayscale values in the low-light inversion image.

[0051] Furthermore, the final enhancement result obtained by inverting the descattered image obtained in step (3) is:

[0052]

[0053] Among them, Output is the desired final enhancement result.

[0054] Beneficial effects: Compared with the existing technology, the present invention has the following beneficial effects: the present invention applies the polarization descattering model to the field of low-light image enhancement, which can simply and efficiently use polarization information to achieve low-light image enhancement of targets with complex polarization characteristics; and does not require human-computer interaction, thereby expanding the application scope of polarization imaging technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is a flow chart of the present invention;

[0056] Figure 2 This is a model diagram of the polarization imaging system in a low-light environment of the present invention;

[0057] Figure 3 Schematic diagram of the constraint decomposition model of the present invention;

[0058] Figure 4 These are images under normal lighting, low lighting, and after adopting the present invention; among them, (a) is the normal lighting image; (b) is the low lighting image; and (c) is the effect diagram of the actual application of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described in detail below with reference to the accompanying drawings.

[0060] like Figure 1 As shown, the present invention proposes a low-light image enhancement method based on a polarization descattering model, comprising the following steps:

[0061] Step 1: Acquire multiple low-light images with different polarization directions, establish a maximum light intensity model and a minimum light intensity model to obtain a maximum light intensity image and a minimum light intensity image.

[0062] like Figure 2 The figure shows a polarization imaging system for low-light environments according to the present invention. An initial beam is emitted by an LED red light source 1, which then illuminates a target 2. Target 2 is composed of plastic and metal materials. In this embodiment, a metal ruler with a plastic sticker attached is used. The high-polarization metal ruler and the low-polarization plastic sticker together form a complex polarization target. Target 2 reflects the initial beam, which then passes through a polarizer 3 and is then illuminated by a CMOS detector 4.

[0063] The polarization direction of the polarizer 3 is rotated to 0° to obtain a polarization image I(0); the polarization direction of the polarizer 3 is rotated to 45° to obtain a polarization image I(45); the polarization direction of the polarizer 3 is rotated to 90° to obtain a polarization image I(90); the polarization direction of the polarizer 3 is rotated to 135° to obtain a polarization image I(135).

[0064] Calculate the Stokes vector using the acquired polarization image:

[0065]

[0066] Where I represents the total light intensity in the low-light scene; Q is the intensity difference between the horizontal and vertical directions; and U is the intensity difference between the 45° and 135° directions.

[0067] Calculate the overall degree of polarization:

[0068]

[0069] Wherein, DoP represents the overall degree of polarization.

[0070] Establish maximum light intensity model and minimum light intensity model:

[0071]

[0072] Among them, I max is the maximum light intensity image; I min is the minimum light intensity image.

[0073] Step 2: Use the maximum light intensity image and the minimum light intensity image to establish a constrained decomposition model to decompose the low-light inversion image, and calculate the decomposed components based on the image enhancement metric.

[0074] With the help of the maximum light intensity image I max With the minimum light intensity image I min Establish a constrained decomposition model to decompose the low-light inverted image:

[0075] R=1-I=(XI max )+(YI min )=R max +R min (4)

[0076] Among them, R is the low light inverted image; X and Y are the values corresponding to I max with I min The decomposition coefficient of , and satisfy the relationship X+Y=1; R max For I max The corresponding decomposition component is the polarization component of the maximum light intensity of the low-light inverted image; R min For I min The corresponding decomposition component is the minimum light intensity polarization component of the low-light inverted image.

[0077] like Figure 3 The figure shows a schematic diagram of the constraint decomposition model of the present invention. X and Y must satisfy the following constraints, namely:

[0078]

[0079] Where ΔI is I max with I min The difference is the low-light polarization difference image.

[0080] Decomposed component R max With R min The decomposition coefficients X and Y need to be used as independent variables for calculation. The method for obtaining the optimal solution that satisfies the maximum image enhancement metric index is as follows: the image enhancement metric index is used as the objective function, that is:

[0081] (X,Y)optimal =argmax{EME(ΔR)} (6)

[0082] Where ΔR is R max With R min The difference between the two images is the low-light inverted polarization difference image; EME(ΔR) is the image enhancement metric of ΔR.

[0083] Specifically, the image enhancement metric is obtained by the following formula:

[0084]

[0085] The image ΔR is divided into k1×k2 blocks numbered (k, l); i max,k,l and i min,k,l is the maximum and minimum value in each block numbered (k, l); q is a very small number that is not divisible by 0, and in this embodiment is 0.001. Specifically, the image enhancement metric EME (ΔR) calculated in this embodiment is a maximum of 0.824, and finally (X, Y) is given. optimal The solution is (0.704, 0.296).

[0086] Step 3: Integrate the obtained components into the scattering degradation physical model, establish a descattering image inversion model, and invert the obtained descattering image to obtain the final enhancement result.

[0087] Introduce the scattering degradation physical model, namely:

[0088] R=D+B=L·t+A ∞ (1-t) (8)

[0089] Where R is the low-light inversion image, which is similar to the scattering degraded image; D is the direct transmitted light; B is the backscattered light; L is the descattered image; t is the medium transmittance; A ∞ is the backscattered light intensity at infinity.

[0090] According to the previous polarization descattering method, the backscattered light B can be expressed as follows:

[0091]

[0092] Wherein, ΔR is the difference between the components obtained in step (2); P scat is the polarization degree of backscattered light, which can be globally estimated through low-pass filtering technology and the definition of polarization degree.

[0093] Combining formulas (8) and (9), the descattered image inversion model is established:

[0094]

[0095] Where L is the descattered image; A ∞ is the backscattered light intensity at infinity, which can be set to the average value of the top 0.1% pixels with the largest grayscale value in the low-light inversion image; in this embodiment, A ∞ is 0.992; P scat is the polarization degree of the backscattered light; ε is a correction coefficient slightly larger than 1 to ensure that B<R. In this embodiment, ε is 1.05.

[0096] Backscattered light polarization degree P scat The global estimation method is: the maximum light intensity polarization component R of the acquired low-light inversion image max The minimum intensity polarization component R of the low-light inverted image min Perform low-pass filtering, that is:

[0097]

[0098] Among them, B max With B min R max With R min The backscattered light estimated in LPF{·} is low-pass filtering. Specifically, this embodiment adopts Gaussian low-pass filtering.

[0099] According to the definition of polarization degree, the polarization degree of backscattered light is estimated globally, namely:

[0100]

[0101] in, is the globally estimated degree of polarization of the backscattered light.

[0102] Substitute the above-obtained parameters into the descattering image inversion model of formula (10) to obtain the corresponding descattering image, and invert the obtained descattering image to obtain the final enhancement result, namely:

[0103]

[0104] Among them, Output is the desired final enhancement result.

[0105] In order to verify the effectiveness of the present invention, target images taken under normal lighting conditions were obtained as references, such as Figure 4 As shown in (a); Figure 4 Figure (b) shows an image of a target captured in low light. The image is not bright enough to the naked eye, and details are hidden in darkness. However, after processing using the method of the present invention, the image brightness is significantly improved, and details are clearly visible, as shown in Figure 4(c). Furthermore, the method of the present invention is not restricted by human-computer interaction and can simply and efficiently utilize polarization information to enhance low-light images of targets with complex polarization characteristics.

[0106] In order to quantitatively evaluate the image quality, the peak signal-to-noise ratio and structural similarity are used to evaluate the quality of the enhanced image. The larger the value, the higher the image quality. The results are shown in Table 1:

[0107] Table 1 Quality evaluation of restored images using peak signal-to-noise ratio and structural similarity

[0108] contrast Peak signal-to-noise ratio Structural similarity Low-light images 7.068 0.154 The present invention enhances the image 21.122 0.760

[0109] As can be seen from Table 1, compared with the original low-light image, the two objective evaluation indicators of the enhanced image of the present invention are significantly improved, verifying the superior enhancement effect of the present invention.

[0110] The present invention has been described in detail above with reference to specific embodiments. However, these descriptions should not be construed as limiting the present invention. Those skilled in the art will appreciate that various equivalent substitutions, modifications, or improvements may be made to the technical solutions and implementations of the present invention without departing from the spirit and scope of the present invention, all of which fall within the scope of the present invention. The scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A low-light image enhancement method based on a polarization descattering model, characterized in that: The following steps are involved: (1) Acquire multiple low-light images with different polarization directions, establish a maximum light intensity model and a minimum light intensity model to obtain a maximum light intensity image and a minimum light intensity image; (2) Using the maximum light intensity image and the minimum light intensity image, a constrained decomposition model is established to decompose the low-light inversion image, and the decomposed components are calculated based on the image enhancement metrics; (3) The decomposed components are integrated into the scattering degradation physical model, a descattering image inversion model is established, and the descattering image is inverted to obtain the final enhancement result.

2. The low-light image enhancement method based on the polarization descattering model according to claim 1, characterized in that: The multiple low-light images with different polarization directions in step (1) are polarization images I(0), I(45), I(90) and I(135) corresponding to polarization directions of 0°, 45°, 90° and 135°.

3. The low-light image enhancement method based on the polarization descattering model according to claim 1, characterized in that: The implementation process of establishing the maximum light intensity model and the minimum light intensity model in step (1) is as follows: Calculate the Stokes vector using the polarization image: Where I represents the total light intensity in the low-light scene; Q is the intensity difference between the horizontal and vertical directions; U is the intensity difference between the 45° and 135° directions; Calculate the overall degree of polarization: Where DoP represents the overall degree of polarization; Establish maximum light intensity model and minimum light intensity model: Among them, I max is the maximum light intensity image; I min is the minimum light intensity image.

4. The low-light image enhancement method based on the polarization descattering model according to claim 1, characterized in that: The implementation process of step (2) is as follows: With the help of the maximum light intensity image I max With the minimum light intensity image I min Establish a constrained decomposition model to decompose the low-light inverted image: R=1-I=(X-I max )+(Y-I min )=R max +R min (4) Among them, R is the low light inverted image; X and Y are the values corresponding to I max with I min The decomposition coefficient of , and satisfy the relationship X+Y=1; R max For I max The corresponding decomposition component is the polarization component of the maximum light intensity of the low-light inverted image; R min For I min The corresponding decomposition component is the minimum light intensity polarization component of the low-light inverted image; The constraint decomposition model must meet the following constraints, namely: Where ΔI is I max with I min The difference is the low-light polarization difference image; Decomposed component R max With R min It is necessary to use the decomposition coefficients X and Y as independent variables for calculation, and use the image enhancement metric as the objective function to obtain the optimal solution that satisfies the maximum image enhancement metric, that is: (X,Y) optimal =argmax{EME(ΔR)} (6) Where ΔR is R max With R min The difference between the two images is the low-light inverted polarization difference image; EME(ΔR) is the image enhancement metric of ΔR.

5. The low-light image enhancement method based on the polarization descattering model according to claim 4, characterized in that: The image enhancement metric EME(ΔR) is: The image ΔR is divided into k1×k2 blocks numbered (k, l); i max,k,l and i min,k,l are the maximum and minimum values in each block numbered (k, l); q is a very small number that avoids being divided by 0.

6. The low-light image enhancement method based on the polarization descattering model according to claim 1, characterized in that: The implementation process of establishing the descattered image inversion model in step (3) is as follows: Introduce the scattering degradation physical model, namely: R=D+B=L·t+A ∞ (1-t) (8) Where R is the low-light inversion image, which is similar to the scattering degraded image; D is the direct transmitted light; B is the backscattered light; L is the descattered image; t is the medium transmittance; A ∞ is the backscattered light intensity at infinity; According to the polarization descattering method, the backscattered light B is: Where ΔR is the difference between the components; P scat is the polarization degree of backscattered light; Combining formulas (8) and (9), the descattered image inversion model is established: Where L is the descattered image; A ∞ is the backscattered light intensity at infinity; P scat is the polarization degree of backscattered light; ε is the correction coefficient to ensure B<R.

7. The low-light image enhancement method based on the polarization descattering model according to claim 6, characterized in that: The backscattered light polarization degree P scat Obtained by global estimation method: The maximum intensity polarization component R of the acquired low-light inversion image max The minimum intensity polarization component R of the low-light inverted image min Perform low-pass filtering, that is: Among them, B max With B min R max With R min The estimated backscattered light in ; LPF{·} is low-pass filtering; According to the definition of polarization degree, the polarization degree of backscattered light is estimated globally, namely: in, is the globally estimated degree of polarization of the backscattered light.

8. The low-light image enhancement method based on the polarization descattering model according to claim 6, characterized in that: The backscattered light intensity at infinity is an average value of the first 0.1% pixels with the largest grayscale value in the low-light inversion image.

9. The low-light image enhancement method based on the polarization descattering model according to claim 1, characterized in that: The final enhancement result obtained by inverting the descattered image obtained in step (3) is: Among them, Output is the desired final enhancement result.

Images