CN107169942A - A kind of underwater picture Enhancement Method based on fish retinal mechanisms - Google Patents

Abstract

The invention discloses an underwater image enhancement method based on the fish retina mechanism. The method of the invention simulates the feedback relationship between the fish retina horizontal cells and the cone cells to remove the color cast of the underwater image, and simulates the fish Center-periphery antagonism of retinal bipolar cells to deblur underwater images. During the whole simulation process, the structure of bipolar cells inhibited by the horizontal cell side of the fish retina was simulated to design the double Gaussian difference filter of the receptive field of bipolar cells; at the same time, the sigmoid curve was used to simulate the continuous release of dopamine regulation by reticulum cells in the dark The activities of horizontal cells make the processed image more in line with the visual mechanism of fish; finally, gamma transformation is used to simulate the nonlinear processing of brightness information by amacrine cells, and constitute the central input of color bipolar cells.

Description

An underwater image enhancement method based on fish retinal mechanism

technical field

The invention belongs to the technical field of image processing and relates to a color image enhancement technology, in particular to an underwater image enhancement method based on a fish retina mechanism.

Background technique

With the continuous enhancement of human exploration capabilities, more and more underwater images have been widely disseminated and applied. However, due to the backscattering and forward scattering of suspended particles in the water body, the image will be blurred, and the color of the underwater image will have a blue-green cast due to the different attenuation speeds of light waves of different wavelengths after the light enters the water. Image blur and color cast will make the final underwater image not clear enough. Therefore, how to remove the influence of blur and color cast and obtain underwater images with high contrast has become a more important issue.

Existing image deblurring methods are mainly methods based on dark channel prior assumptions. These methods are generally based on atmospheric scattering physical models. Representative methods include the method proposed by Chiang J Y and others in 2012. References: ChiangJ Y, Chen Y C. Underwater image enhancement by wavelength compensation and dehazing [J]. IEEE Transactions on Image Processing, 2012, 21(4): 1756-1769. They all need to satisfy the dark channel prior to achieve a good deblurring effect.

Color constancy methods include learning-based or static color constancy methods, which mainly restore the true color of objects by estimating the color of the scene light source. However, they are mainly aimed at land scenes, ignoring some characteristics of underwater images. Among them, the learning-based method is still difficult to apply to underwater image processing because there is currently no underwater image database with a standard light source; while the static method is mostly based on certain grayscale assumptions, but generally the red band of underwater images is obvious. It is weaker than other bands and does not meet the grayscale assumption, so the color of the image corrected by the static method is mostly reddish. Even some improved static methods must meet the dark channel prior model to have a certain effect. Fish retina can solve the problem of color cast and blur at the same time, but there is no method to simulate fish retina to deal with color cast and blur at the same time.

Contents of the invention

In view of the above-mentioned problems in the prior art, the present invention proposes an underwater image enhancement method based on the fish retina mechanism.

The technical scheme of the present invention is: a kind of underwater image enhancement method based on fish retinal mechanism, comprises the following steps:

S1. Extract color components and brightness components: extract the red component I _R , green component I _G , and blue component I _B for each pixel of the underwater image, and calculate the average brightness component I:

I＝(I _R +I _G +I _B )/3

S2. Calculate the adjusted mean value of the three RGB channels: calculate the mean value M _r of the pixels in the brightest part of the red channel whose pixel value is greater than the first threshold, as the adjusted mean value of the red channel, and calculate the green channel and the blue channel Respective mean values M _g , M _b ;

S3. Correct the color shift of the image: divide each pixel of the three channels R, G, and B by its corresponding mean value, and obtain the updated values I′ _R , I′ _G , and I′ _B of each channel after the processing is completed, and calculate in detail The formula is;

Then, stretch the updated value to the original image brightness, the specific calculation formula is:

Among them, I' represents the image composed of I' _R , I' _G , and I'_B; mean represents the mean value of the image.

S4, calculate color channel and luminance channel to experience field peripheral input: obtain R, G, the value (I″ R , I″ G , I″ B ) of the luminance component I that step S1 obtains and step S3 three channels update (I″ _R , I″ _G , I″ _B ) respectively Perform filtering to obtain the peripheral input f _sI , f _sR , f _sG , f _sB of the four channels;

S5. Calculate the input of the receptive field center of the brightness channel:

Calculate the mean value M of the luminance channel I obtained in step S1, if M is less than the second threshold, then the center input f _cI of the receptive field of the luminance channel is adjusted by a sigmoid function, and meanwhile, the (I″ _R , I″ _G obtained in step S3 , I″ _B ) also use sigmoid to update again; otherwise let f _cI =I, and (I″ _R , I″ _G , I″ _B ) do not update;

S6. Calculate the weight of the color channel and the brightness channel to experience the field week: use k to represent the weight of the RGB channel and the brightness channel to experience the field week, and the calculation formula is:

Among them, λ represents the three channels of R, G, and B, and A is the maximum value corresponding to each channel. I″ _λ (x, y) is the pixel value corresponding to the (x, y) position of I″ _R , I″ _G , and I″ _B after the processing in step S5, and k _MAX is the upper limit of k value.

S7. Calculate the receptive field response of the luminance channel: Substitute the receptive field center and peripheral inputs f _cI and f _sI calculated in steps S4 and S5 into the double Gaussian difference function to calculate the receptive field response value of the luminance channel. The specific calculation formula is:

in, Represents convolution, f _cI (x, y), f _sI (x, y) represent the receptive field center and peripheral input of the point (x, y) in the image, g(m, n; σ _c ), g(m, n; σ _s ) represents a two-dimensional Gaussian function with a size of m*n, and roddB _p is the output result of the receptive field of the luminance channel.

S8. Calculate the central input of the RGB three-channel receptive field: perform gamma transformation on the luminance channel receptive field output roddB _p obtained in step S7 to obtain roddB _p ^γ , and compare it with the I″ _R , I″ _G , and I″ obtained through the processing of step S5 Multiplying _B together constitutes the input f _c of the receptive field center of the three channels R, G, and B. The specific calculation formula is:

f _cR ＝I″ _R *rodB _p ^γ

f _cG ＝I″ _G *rodB _p ^γ

f _cB ＝I″ _B *rodB _p ^γ

Among them, * represents the multiplication sign;

S9. Calculate and output RGB three-channel receptive field response: Same as step S7, calculate the receptive field center input f _cR , f _cG , f _cB and peripheral input f _sR , f of R, G, and B channels in steps S5 and S8 _{Substituting sG} and f _sB into the double Gaussian difference function to calculate the receptive field response of the R, G, and B channels, the specific calculation formula is:

The receptive field responses B _pR , B _pG , and B _pB of the three channels R, G, and B are the enhanced dehazed images of the three channels, and the three channels are recombined into an RGB image as the final output.

Further, the first threshold in step S2 is 0.1, and the second threshold in step S5 is 0.5.

Further, the brightest part of pixels mentioned in step S2 is specifically 50% of the brightest pixels.

Further, the filtering described in step S4 is specifically mean filtering.

Further, in the step S2, in order to avoid the adjusted mean value of the red channel being too high, when the adjusted red channel mean value _Mr is greater than the green channel mean value _Mg , the green channel mean value _Mg is used as the red channel after final adjustment. mean, that is:

M _r =min(M _r , M _g ).

Further, in step S5, the central input f _cI of the receptive field of the luminance channel is adjusted using a sigmoid function as follows:

Furthermore, the window width of the mean filter described in step S4 is any size larger than 3*3 and smaller than 15*15, such as 7*7, 9*9 and so on.

Further, the Gaussian functions of the receptive field center and the periphery described in step S7 and step S9 are specifically:

Wherein, the value range of δ _c is specifically 0.2-0.8, the value of δ _s is 3 times of δ _c , and the value range of m, n is specifically an integer of 5-15.

Further, the value range of γ in step S8 is specifically 0.4-0.6.

The beneficial effects of the present invention are: the method of the present invention simulates the feedback relationship between fish retinal horizontal cells and cone cells to remove the color cast of underwater images, and simulates the central peripheral antagonism of fish retinal bipolar cells to remove Blurring of underwater images. During the whole simulation process, the structure of bipolar cells inhibited by the horizontal cell side of the fish retina was simulated to design the double Gaussian difference filter of the receptive field of bipolar cells; at the same time, the sigmoid curve was used to simulate the continuous release of dopamine regulation by reticulum cells in the dark The activities of horizontal cells make the processed image more in line with the visual mechanism of fish; finally, gamma transformation is used to simulate the nonlinear processing of brightness information by amacrine cells, and constitute the central input of color bipolar cells. The algorithm based on the invention can be embedded in the camera as an underwater mode to deal with the color shift and blur problems of underwater images.

Description of drawings

FIG. 1 is a flowchart of underwater image processing according to an embodiment of the present invention.

Figure 2 is the original image taken underwater with color cast and blur problems.

Figure 3 is the corresponding result after removing the color shift from the original image.

Figure 4 shows the corresponding image after the original image has been updated twice.

Figure 5 is the response image of the receptive field of the luma channel.

Figure 6 is the final output image with color cast and blur removed.

detailed description

Embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

Fish have strong adaptability to color cast and blur in underwater images. Learning the image processing process of fish visual system can help solve the problems of color cast and blur in underwater images captured by cameras. Based on this, the present invention proposes an underwater image enhancement method based on the fish retina mechanism, as shown in Figure 1, comprising the following steps:

Be 768*1024 for an image size, have the underwater image (as shown in Figure 2) of color cast and fuzzy problem, the detailed step flow process of the present invention is as follows:

S1. Extract color components and brightness components: extract red component I _R , green component I _G , and blue component I _B for each pixel of the underwater image, and calculate the average brightness component I.

Take example point 1 with pixel values (0.659, 0.718, 0.463) and example point 2 with pixel values (0.275, 0.373, 0.212) in the original input image (Fig. 2). Their corresponding average brightness components I are (0.659+0.718+0.463)/3=0.613 and (0.275+0.373+0.212)/3=0.286, respectively.

S2. Calculating the adjusted average of the three RGB channels: calculating the average of the three channels respectively, wherein the R channel calculates the brightest 50% of the pixels greater than 0.1. In the original image (Fig. 2), the mean value of the first 50% pixels with the R channel pixel value greater than 0.1 is 0.4231, so M _r =0.4231, the G channel mean value is 0.5407, so M _g =0.5407, and the B channel mean value is 0.3367, so M _b = 0.3367. Since the mean value of the R channel is smaller than the mean value of the G channel, the mean value M _r of the R channel remains unchanged at 0.4231.

It should be noted that, unlike calculating the average value of the R channel, here a general method is used to calculate the average value of the G and B channels.

S3. Correct the color shift of the image: divide each pixel of the three channels R, G, and B by its corresponding mean value, and obtain the updated values I′ _R , I′ _G , and I′ _B of each channel after the processing is completed, and update them , the image color cast is removed.

After dividing the pixels of the channels corresponding to the two example pixels by the mean value, the updated pixel values I′ _R , I′ _G , and I′ _B are respectively (0.659/0.4231, 0.718/0.5407, 0.463/0.3367)=(1.5576 , 1.3279, 1.3751) and (0.275/0.4231, 0.373/0.5407, 0.212/0.3367) = (0.6500, 0.6898, 0.6296).

At this time, the mean (I') of the image I' composed of I' _R , I' _G , and I' _B is calculated to be 0.9985, and the mean (I) of the original image brightness is 0.4170. Therefore, the example point 1 The I′ _R , I′ _G , and I′ _B stretched to the brightness of the original image become:

Similarly, the example point 2 is stretched to the corresponding I″ _R , I″ _G , and I″ _B of the brightness of the original image to be (0.2714, 0.2881, 0.2630) respectively. Through this update, the color cast of the original image is removed, as shown in Figure 3 The corresponding image after the color cast is removed, it can be seen that the green color cast of the image is effectively removed.

S4. Calculate the color channel and the brightness channel to experience the peripheral input: the brightness channel I obtained by S1 and the updated values I″ _R , I″ _G , and I″ _B of the R, G, and B three channels obtained by S3 are respectively subjected to mean value filtering, Peripheral inputs f _sI , f _sR , f _sG , f _sB of the four channels are obtained.

In this embodiment, taking the average filter with a window width of 9*9 as an example, the brightness image I obtained by S1 is average-filtered, and the values of the corresponding positions of the two example pixel points after filtering are obtained as f _sI are 0.7839 and 0.1327 respectively . The images I″ _R , I″ _G , and I″ _B of the updated RGB three channels obtained by S3 are subjected to mean value filtering, and f _sR , f _sG , and f sB of the corresponding positions of the two example points are respectively ( _0.6597 , 0.7313, 0.7448) and (0.2930, 0.1926, 0.1304).

S5, calculate the center input of the brightness channel receptive field: calculate the mean value of the brightness channel, the mean value M of the brightness channel I corresponding to the original input image (Fig. 2) is 0.4170, because it is less than 0.5, so use the sigmoid function:

Computes the center input f _cI of the receptive field of the luma channel. Substitute the luminance I 0.613 and 0.286 of the two example points calculated by S1 into the above formula, and calculate the receptive field center input f _cI of the luminance channel to be 0.7559 and 0.1053 respectively.

At the same time, the values (0.6505, 0.5546, 0.5743) of I″ _R , I″ _G , and I″ _B calculated by S3 are processed by the same function, and two example pixels I″ _R , I″ _G , I″ are substituted into ″ The updated values of _B are (0.8183, 0.6332, 0.6777) and (0.0924, 0.1073, 0.0855), respectively.

FIG. 4 shows the corresponding images after I″ _R , I″ _G , and I″ _B are updated again.

S6. Calculate the weight of the color channel and the brightness channel to experience the field week: use k to represent the weight of the RGB channel and the brightness channel to experience the field week, and the calculation formula is:

Here, in order to avoid the phenomenon that the image is over-enhanced, a reasonable upper limit should be set for the value of k. In this embodiment, this upper limit is set to 0.4, that is, k _MAX =0.4, λ represents three channels of R, G, and B, and A is the maximum value corresponding to each channel. I' _λ (x, y) is the pixel value corresponding to (x, y) position of I " _R , I " _G , I " _B after step S5 processing. At this moment, the maximum value A _λ of RGB three channels is respectively: 0.9465, 0.8778, 0.9333, RGB three channels The calculation results are: 0.2103, 0.2066, 0.1432, so Since 0.2151<0.4, the value of k is 0.2151.

S7. Calculating the receptive field response of the luminance channel: Substituting the receptive field center and peripheral inputs f _cI and f _sI of the luminance channel calculated in steps S4 and S5 into the double Gaussian difference function to calculate the receptive field response value of the luminance channel. The calculation formula is:

In this embodiment, taking σ _c =0.5, σ _s =1.5, m=n=9 as an example, the receptive field response values roddB _p of the luminance channel of two example points are calculated to be 0.4745 and 0.2608 respectively.

Figure 5 shows the response plot of the receptive field for the luma channel.

S8, calculate the central input of the RGB three-channel receptive field: perform gamma transformation on the brightness channel receptive field output rod dB _p obtained in S7, and form R and G together with the I″ _R , I″ _G , and I″ _B obtained through the processing of step S5 , Input f _c of the receptive field center of the three channels B.

In the gamma transformation described in this embodiment, γ=0.5 is taken as an example, and the center inputs f _cR , f _cG , and f _cB of the RGB three-channel receptive fields are calculated. Then the calculation process and result corresponding to example point 1 are:

f _cB ＝I″ _R *rodB _p ^γ ＝0.8183*0.4745 ^0.5 ＝0.5637

f _cG ＝I″ _G *rodB _p ^γ ＝0.6332*0.4745 ^0.5 ＝0.4363

f _cB ＝I″ _B *rodB _p ^γ ＝0.6777*0.4745 ^0.5 ＝0.4669

Similarly, the center input f _cR , f _cG , and f _cB of the receptive field corresponding to example point 2 are calculated as (0.0637, 0.0739, 0.0589).

S9. Calculating and outputting the receptive field responses of the RGB three-channels: Same as step S7, in this embodiment, the double Gaussian difference function is used to calculate the receptive field responses of the R, G, and B channels:

Substitute f _cR , f _cG , f _cB calculated in S8 and k calculated in S6 into the above formula, σ _c and σ _s adopt the same values as in step S7 σ _c =0.5, σ _s =1.5, m=n= 9. Obtain the three-channel receptive field responses B _pR , B _pG , and B _pB . The calculation results of B _pR , B _pG , and B _pB at the corresponding positions of the two example points are (1,0.778,0.865) and (0.022,0.0184,0.0376) respectively. Finally, output the calculated results.

Figure 6 shows the final output image. Compared with the original image (Figure 2), the color cast and blur of the image are effectively removed.

The above simple examples are mainly explained by taking a single pixel value of an image as an example, and the actual calculation is performed on all pixels of the entire image. Through such a simple example, the whole process of simulating fish retina removing color shift and blurring in the present invention is illustrated.

The embodiments described here are to help readers understand the principles of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical revelations disclosed in the present invention without departing from the essence of the present invention, and these modifications and combinations are still within the protection scope of the present invention.

Claims (9)

1. An underwater image enhancement method based on a fish retina mechanism comprises the following steps:

s1, extracting color component and brightness component: respectively extracting red component I from each pixel point of underwater image_RGreen component I_GBlue component I_BAnd calculates the average luminance component I:

I＝(I_R+I_G+I_B)/3

s2, calculating the adjusted mean value of the RGB three channels: calculating the brightest pixel value in the red channel that is greater than a first thresholdAverage value M of partial pixel points_rThe average value M of the green channel and the blue channel is calculated as the adjusted average value of the red channel_g、M_b；

S3, correcting color cast of the image: dividing each pixel point of R, G, B three channels by the corresponding mean value thereof, and obtaining the updated value I 'of each channel after the processing is finished'_R、I′_G、I′_BThe specific calculation formula is as follows;

<mrow> <msubsup> <mi>I</mi> <mi>R</mi> <mo>&amp;prime;</mo> </msubsup> <mo>=</mo> <mfrac> <msub> <mi>I</mi> <mi>R</mi> </msub> <msub> <mi>M</mi> <mi>r</mi> </msub> </mfrac> </mrow>

<mrow> <msubsup> <mi>I</mi> <mi>G</mi> <mo>&amp;prime;</mo> </msubsup> <mo>=</mo> <mfrac> <msub> <mi>I</mi> <mi>G</mi> </msub> <msub> <mi>M</mi> <mi>g</mi> </msub> </mfrac> </mrow>

<mrow> <msubsup> <mi>I</mi> <mi>B</mi> <mo>&amp;prime;</mo> </msubsup> <mo>=</mo> <mfrac> <msub> <mi>I</mi> <mi>B</mi> </msub> <msub> <mi>M</mi> <mi>b</mi> </msub> </mfrac> </mrow>

and then stretching the updated value to the brightness of the original image, wherein the specific calculation formula is as follows:

<mrow> <msubsup> <mi>I</mi> <mi>R</mi> <mrow> <mo>&amp;prime;</mo> <mo>&amp;prime;</mo> </mrow> </msubsup> <mo>=</mo> <mfrac> <mrow> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> <mrow> <mo>(</mo> <mi>I</mi> <mo>)</mo> </mrow> <mo>*</mo> <msubsup> <mi>I</mi> <mi>R</mi> <mo>&amp;prime;</mo> </msubsup> </mrow> <mrow> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> <mrow> <mo>(</mo> <msup> <mi>I</mi> <mo>&amp;prime;</mo> </msup> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>

<mrow> <msubsup> <mi>I</mi> <mi>G</mi> <mrow> <mo>&amp;prime;</mo> <mo>&amp;prime;</mo> </mrow> </msubsup> <mo>=</mo> <mfrac> <mrow> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> <mrow> <mo>(</mo> <mi>I</mi> <mo>)</mo> </mrow> <mo>*</mo> <msubsup> <mi>I</mi> <mi>G</mi> <mo>&amp;prime;</mo> </msubsup> </mrow> <mrow> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> <mrow> <mo>(</mo> <msup> <mi>I</mi> <mo>&amp;prime;</mo> </msup> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>

<mrow> <msubsup> <mi>I</mi> <mi>B</mi> <mrow> <mo>&amp;prime;</mo> <mo>&amp;prime;</mo> </mrow> </msubsup> <mo>=</mo> <mfrac> <mrow> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> <mrow> <mo>(</mo> <mi>I</mi> <mo>)</mo> </mrow> <mo>*</mo> <msubsup> <mi>I</mi> <mi>B</mi> <mo>&amp;prime;</mo> </msubsup> </mrow> <mrow> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> <mrow> <mo>(</mo> <msup> <mi>I</mi> <mo>&amp;prime;</mo> </msup> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>

wherein I 'represents I'_R、I′_G、I′_BForming an image, wherein mean represents the mean value of the image;

s4, calculating color channel and brightness channel, and sensing field input: for step S1The obtained luminance component I and the step S3 obtain R, G, B three channels of updated values (I ″)_R、I″_G、I″_B) Respectively filtering to obtain peripheral input f of the receptive fields of the four channels_sI、f_sR、f_sG、f_sB；

S5, calculating the center input of the receptive field of the brightness channel:

calculating the average value M of the brightness channel I obtained in the step S1, and if M is smaller than a second threshold value, inputting the center f of the brightness channel receptive field_cIAdopting sigmoid function to adjust, and simultaneously, obtaining (I') obtained in step S3_R、I″_G、I″_B) Updating again by adopting sigmoid; otherwise make f_cIIs ═ I, and (I ″)_R、I″_G、I″_B) Updating is not carried out;

s6, calculating the weight occupied by the color channel and the brightness channel in the field experience period: k represents the RGB channel and brightness channel sensing field week weight, and the calculation formula is as follows:

<mrow> <mi>k</mi> <mo>=</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <msub> <mi>k</mi> <mrow> <mi>M</mi> <mi>A</mi> <mi>X</mi> </mrow> </msub> <mo>,</mo> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> <mo>(</mo> <mrow> <munder> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <mi>&amp;lambda;</mi> </munder> <mfrac> <mrow> <msubsup> <mi>I</mi> <mi>&amp;lambda;</mi> <mrow> <mo>&amp;prime;</mo> <mo>&amp;prime;</mo> </mrow> </msubsup> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> </mrow> <msub> <mi>A</mi> <mi>&amp;lambda;</mi> </msub> </mfrac> </mrow> <mo>)</mo> <mo>)</mo> </mrow> </mrow>

where λ represents R, G, B three channels, and a is the maximum value for each channel. I ″)_λ(x, y) is I ″, which is processed in step S5_R、I″_G、I″_BPixel value, k, corresponding to the (x, y) position_MAXThe upper limit of the k value.

S7, calculating the response of the brightness channel receptive field: inputting the central and peripheral receptive field inputs f calculated in steps S4 and S5_cIAnd f_sISubstituting the double Gaussian difference function to calculate the receptive field response value of the brightness channel, wherein the specific calculation formula is as follows:

<mrow> <msub> <mi>rodB</mi> <mi>p</mi> </msub> <mo>=</mo> <mi>m</mi> <mi>a</mi> <mi>x</mi> <mo>&amp;lsqb;</mo> <mn>0</mn> <mo>,</mo> <msub> <mi>f</mi> <mrow> <mi>c</mi> <mi>I</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>&amp;CircleTimes;</mo> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mi>k</mi> <mo>&amp;CenterDot;</mo> <msub> <mi>f</mi> <mrow> <mi>s</mi> <mi>I</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>n</mi> <mo>,</mo> <mi>n</mi> <mo>)</mo> </mrow> <mo>&amp;CircleTimes;</mo> <mi>g</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>s</mi> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> </mrow>

wherein,representing a convolution, f_cI(x，y)、f_sI(x, y) represents the receptive field center and peripheral inputs for point (x, y) in the image, g (m, n;σ_c)、g(m，n；σ_s) Representing a two-dimensional Gaussian function of size m x n, rodB_pNamely the receptive field output result of the brightness channel.

S8, calculating the central input of RGB three-channel receptive field: outputting rodB to the brightness channel receptive field obtained in the step S7_pGamma conversion is carried out to obtain rodB_p ^γAnd is compared with the I' processed in step S5_R、I″_G、I″_BThe multiplication jointly forms the central input f of the receptive field of R, G, B three channels_cThe specific calculation formula is as follows:

f_cR＝I″_R*rodB_p ^γ

f_cG＝I″_G*rodB_p ^γ

f_cB＝I″_B*rodB_p ^γ

wherein denotes a multiplication number;

s9, calculating RGB three-channel receptive field response and outputting: in the same step S7, the central input f of the receptor field of R, G, B three channels is obtained by calculating the steps S5 and S8_cR、f_cG、f_cBAnd peripheral input f_sR、f_sG、f_sBSubstituting the double Gaussian difference function to calculate the receptive field response of R, G, B three channels, wherein the specific calculation formula is as follows:

<mrow> <msub> <mi>B</mi> <mrow> <mi>p</mi> <mi>R</mi> </mrow> </msub> <mo>=</mo> <mi>m</mi> <mi>a</mi> <mi>x</mi> <mo>&amp;lsqb;</mo> <mn>0</mn> <mo>,</mo> <msub> <mi>f</mi> <mrow> <mi>c</mi> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>&amp;CircleTimes;</mo> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mi>k</mi> <mo>&amp;CenterDot;</mo> <msub> <mi>f</mi> <mrow> <mi>s</mi> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>&amp;CircleTimes;</mo> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>s</mi> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> </mrow>

<mrow> <msub> <mi>B</mi> <mrow> <mi>p</mi> <mi>G</mi> </mrow> </msub> <mo>=</mo> <mi>m</mi> <mi>a</mi> <mi>x</mi> <mo>&amp;lsqb;</mo> <mn>0</mn> <mo>,</mo> <msub> <mi>f</mi> <mrow> <mi>c</mi> <mi>G</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>&amp;CircleTimes;</mo> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mi>k</mi> <mo>&amp;CenterDot;</mo> <msub> <mi>f</mi> <mrow> <mi>s</mi> <mi>G</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>&amp;CircleTimes;</mo> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>s</mi> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> </mrow>

<mrow> <msub> <mi>B</mi> <mrow> <mi>p</mi> <mi>B</mi> </mrow> </msub> <mo>=</mo> <mi>m</mi> <mi>a</mi> <mi>x</mi> <mo>&amp;lsqb;</mo> <mn>0</mn> <mo>,</mo> <msub> <mi>f</mi> <mrow> <mi>c</mi> <mi>B</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>&amp;CircleTimes;</mo> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mi>k</mi> <mo>&amp;CenterDot;</mo> <msub> <mi>f</mi> <mrow> <mi>s</mi> <mi>B</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>&amp;CircleTimes;</mo> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>s</mi> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> </mrow>

r, G, B receptor field response of three channels_pR、B_pG、B_pBNamely, the defogged image after the enhancement of the three channels is recombined into an RGB image as the final output.

2. The underwater image enhancement method based on fish retina mechanism of claim 1, wherein said first threshold of step S2 is 0.1, and said second threshold of step S5 is 0.5.

3. The underwater image enhancement method based on the fish retina mechanism of claim 1, wherein the brightest portion of the pixels in step S2 is specifically the brightest 50% of the pixels.

4. The underwater image enhancement method based on the fish retina mechanism of claim 1, wherein the filtering of step S4 is mean filtering.

5. The underwater image enhancement method based on fish retina mechanism as claimed in claim 1, wherein the underwater image enhancement method is characterized in thatIn step S2, to avoid the adjusted red color channel mean value being too high, the adjusted red color channel mean value M_rGreater than the mean value M of the green channel_gUsing the mean value M of the green channel_gAs the final adjusted mean value of the red channel, namely:

M_r＝min(M_r，M_g)。

6. the underwater image enhancement method based on fish retina mechanism of claim 1, wherein step S5 inputs f to the center of the brightness channel receptive field_cIThe sigmoid function is adopted for regulation specifically as follows:

<mrow> <msub> <mi>f</mi> <mrow> <mi>c</mi> <mi>I</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mo>(</mo> <mn>1</mn> <mo>+</mo> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mn>10</mn> <mrow> <mo>(</mo> <mi>I</mi> <mo>-</mo> <mn>0.5</mn> <mo>)</mo> </mrow> </mrow> </msup> <mo>)</mo> </mrow> </mfrac> <mo>.</mo> </mrow>

7. the underwater image enhancement method based on the fish retina mechanism of claim 4, wherein the window width of the mean filter of step S4 is any size greater than 3 × 3 and less than 15 × 15.

8. The underwater image enhancement method based on fish retina mechanism of claim 1, wherein the gaussian functions of the center and periphery of the receptive field in steps S7 and S9 are specifically:

<mrow> <msub> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mn>2</mn> <msub> <mi>&amp;pi;&amp;sigma;</mi> <mi>c</mi> </msub> </mrow> </mfrac> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mfrac> <mrow> <msup> <mi>m</mi> <mn>2</mn> </msup> <mo>+</mo> <msup> <mi>n</mi> <mn>2</mn> </msup> </mrow> <mrow> <mn>2</mn> <msubsup> <mi>&amp;sigma;</mi> <mi>c</mi> <mn>2</mn> </msubsup> </mrow> </mfrac> </mrow> </msup> </mrow>2

<mrow> <msub> <mi>g</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>,</mo> <mi>n</mi> <mo>;</mo> <msub> <mi>&amp;sigma;</mi> <mi>s</mi> </msub> <mo>)</mo> </mrow> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mn>2</mn> <msub> <mi>&amp;pi;&amp;sigma;</mi> <mi>s</mi> </msub> </mrow> </mfrac> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mfrac> <mrow> <msup> <mi>m</mi> <mn>2</mn> </msup> <mo>+</mo> <msup> <mi>n</mi> <mn>2</mn> </msup> </mrow> <mrow> <mn>2</mn> <msubsup> <mi>&amp;sigma;</mi> <mi>s</mi> <mn>2</mn> </msubsup> </mrow> </mfrac> </mrow> </msup> </mrow>

wherein,_cthe value range of (A) is specifically 0.2-0.8,_stake a value of_cThe value range of m and n is specifically an integer of 5-15.

9. The underwater image enhancement method based on the fish retina mechanism according to claim 1, wherein a value range of γ in step S8 is specifically 0.4-0.6.