CN106981053A - A kind of underwater picture Enhancement Method based on Weighted Fusion - Google Patents

Abstract

The invention discloses an underwater image enhancement method based on weighted fusion. The underwater image enhancement method includes the following steps: the degraded underwater image is equalized by Gray-World and histogram respectively to obtain an input image; Through normalization processing, weighted guided filtering method is used to define the weight factor, modify the weight factor, and obtain the corrected weight map; use the Laplacian pyramid to decompose each input map separately, and each A weight map is decomposed with a Gaussian pyramid, and finally a multi-scale fusion method is used to fuse the input image and the weight map to obtain an image with rich details. This method does not need complex deconvolution operations, and the weights are selected reasonably, so that the image has richer details on the basis of color correction.

Description

A Method of Underwater Image Enhancement Based on Weighted Fusion

technical field

The invention relates to the field of image fusion underwater image enhancement, in particular to an underwater image enhancement method based on weighted fusion.

Background technique

Underwater image enhancement technology is an important part of obtaining marine information, and it is also an important technology for completing underwater operations. Due to the attenuation effect of light transmission in water, red light attenuates the fastest, blue-green light attenuates the slowest, and suspended matter in water will cause multiple scattering of light in water, forward scattering causes image blur, and back scattering causes image blurring. The contrast decreases. Therefore, the underwater image will show a blue-green tone with low contrast and clarity, which will affect the progress of underwater exploration. The existing underwater image enhancement technology is still immature and needs further in-depth exploration.

As early as 1979, McGlamery [1] proposed a classic underwater imaging model. He proposed that the optical radiation received by the imaging system consists of three components: directly attenuated light, forward scattering, and backscattering. On the basis of this theoretical model, the most widely used are the defogging method based on the Dark Channel Prior (DCP) and the enhancement method based on Retinex. In 2008, Fattal [2] used the assumption that the image surface shadow and the atmospheric transfer function are locally uncorrelated to estimate the transmittance of the scene. In 2011, Zhang Kai et al. [3] equated the water body scattering effect to the change of ambient light. Through the multi-scale Retinex algorithm processing under the brightness channel of the underwater color image, the water body scattering effect can be reduced and the image contrast can be improved. However, the algorithm Noise is prone to appear in the background area of the image. In 2012, Chiang[4] proposed an underwater image enhancement algorithm based on wavelength compensation for dehazing. He obtained a depth map considering the effect of artificial light sources, segmented the foreground and background, and compensated the color channels in different proportions, but This method has high computational complexity. In 2013, Hitam et al. [5] enhanced underwater images with a contrast-limited histogram equalization algorithm that combines RGB space and HSV space. This algorithm has high computational efficiency, but it also produces large noise. In 2014, Xueyang Fu[6] et al. proposed an underwater image enhancement method based on the Retinex theory under the variational framework.

In 2012, Ancuti et al. [7] proposed a fusion-based underwater image enhancement algorithm. The algorithm mainly selects the input image and the weight map and achieves the purpose of enhancing the underwater image through multi-resolution fusion. The outstanding features of underwater images are color distortion and contrast reduction. In view of this feature, Ancuti proposes to process each feature separately to obtain two input images and weight images, and then restore the underwater image through multi-resolution fusion. First of all, light of different wavelengths will be absorbed to different degrees during the transmission process, resulting in color shift of the image, and the color constancy algorithm is a correction method for the shifted color. For the above characteristics of underwater images, Ancuti uses the traditional The Gray-World[8] method processes the degraded image to obtain the input image I ¹ , where the illuminance is adjusted:

μ _I ＝0.5+λμ _ref (1)

Among them, μ _ref is the average value of brightness, and μ _I is the estimated value of brightness (the value obtained in Gray-World). After light attenuation, the global contrast of the image is obviously weakened. In order to obtain a clear image, Ancuti uses the traditional histogram equalization method to improve the global contrast and obtain the input image I ² .

After obtaining the two input images with color correction and global contrast improvement, considering that the degraded image still has a lot of deficiencies in saliency, local and global contrast, and exposure, the weight of the input image will be determined by the following four weighting factors:

(1) The Laplacian contrast weight is obtained by applying a Laplacian filter to each luminance channel and calculating its absolute value to obtain a global contrast weight map.

(2) The local contrast weight is obtained from each pixel and its neighboring pixels:

W _LC (x,y)＝||I ^k -I ^k _whc || (2)

Among them, I ^k is the brightness channel of the input image, and I ^k _whc is the channel processed by the low-pass filter.

(3) The main information of the image is only concentrated in a few key areas, and people usually focus on the area where the image contour curvature is the largest or the contour direction changes suddenly. These information are reflected by the saliency map, and the saliency weight map Obtained by Achanta[9].

(4) The exposure weight is used to measure the exposure of the pixel, which is obtained by a Gaussian model:

Wherein, I ^k (x, y) represents the brightness value at (x, y); σ is 0.25.

In order to get good results, this method normalizes the four weight maps to obtain two weight maps, as follows:

Among them, n_W _L i , n_W _LC i , n_W _S i and n_W _E i (i=1, 2) are normalized Laplacian weights, local contrast weights, saliency weights and exposure weights, respectively. Finally, the two input images and the two weight images are processed by the Laplacian pyramid multi-resolution fusion method to obtain an enhanced image:

where L{I} is the Laplacian pyramid of the input image, is a Gaussian pyramid of weights.

references

[1]McGlamery B L.A computer model for underwater camera system[C]//Ocean Optics VI.International Society for Optics and Photonics,1980:221-231.

[2] R. Fattal, "Single Image Dehazing," J.ACM Siggraph 08, 1-9(2008).

[3] Zhang Kai, Qiu Su, Wang Xia. Multi-scale Retinex Enhancement Algorithm for Brightness Channel of Underwater Color Image[J]. Infrared Technology, 2012,33(11):630-634.

[4] Chiang J Y and Chen Ying-Ching. Underwater image enhancement by wavelength compensation and dehazing [J]. IEEE Transactions on Image Processing, 2012, 21(4): 1756-1769.

[5]Hitam M S, Yussof W, Awalludin E A. Mixture contrast limited adaptive histogram equalization for underwater enhancement[C]//International Conference on Computer Applications Technology, Sousse, Tunisia: IEEE Press, 2013:1-5.

[6] X.Y.Fu and P.X.Zhuang, "A Retinex-based Enhancing Approach for Single Underwater Image," IEEE Inter.Conf.Image Process., Paris, France, October 2014, pp.27-30.

[7] C.Ancuti, C.O.Ancuti, T.Haber and P.Bekaert, “Enhancing underwater images and videos by fusion,” in proc.IEEE Conf.Comput.Vis.Patt.Recogni.(CVPR),Providence,RI,Jun .2012, pp.81-88.

[8] B. Gershon, "A spatial processor model for object colourperception," J. Frank. Inst., vol.310, no.1, pp.1-26, 1980.

[9] R. Achantay, S. Hemamiz, F. Estraday, and S. Susstrunky. Frequency-tunedsalient region detection. IEEE CVPR, 2009.

Contents of the invention

The present invention provides an underwater image enhancement method based on weighted fusion. The present invention combines weight factors and Laplacian pyramid fusion. See the following description for details:

An underwater image enhancement method based on weighted fusion, said underwater image enhancement method comprising the following steps:

The degraded underwater image is equalized with Gray-World and histogram respectively to obtain the input image;

Through the normalization process, the method of weighted guided filtering is used to realize the definition of the weight factor, and the weight factor is corrected to obtain the corrected weight map;

Each input image is decomposed by Laplacian Pyramid, and each weight image is decomposed by Gaussian Pyramid. Finally, the input image and weight image are fused by multi-scale fusion method to obtain an image with rich details.

The steps for obtaining the corrected weight map are specifically:

m_W ₁ ＝a_1*n_W _LC 1+b_1*n_W _S 1+c_1*n_W _E 1

m_W ₂ ＝a_2*n_W _LC 2+b_2*n_W _S 2+c_2*n_W _E 2

Among them, a_i, b_i and c_i are the scaling factors of each weight, respectively, n_W _LC i, n_W _S i and n_W _E i are the normalized weight results of the above local contrast weight, saliency weight and exposure weight respectively; i =1,2.

The steps of implementing the definition of the weight factor by using the method of weighted guided filtering are as follows:

Among them, ε is a small positive number to prevent the denominator from being zero; n_W _LC k, n_W _S k and n_W _E k are the normalized weights of local contrast weight, saliency weight and exposure weight respectively; K is 2; N is the total number of pixels.

The beneficial effect of the technical solution provided by the present invention is: adopting the technical solution of the present invention, the overall sensitive effect is better from a subjective point of view, and the brightness is uniform, and the color is more abundant. With good effect. Objectively, the experimental results are measured by four image quality measurement indicators, namely image mean, average gradient, information entropy and standard deviation. Figures 3 to 6 show the results of the objective indicators in Figure 2. From this point of view, the results of the algorithm of the present invention are higher than those of the other three methods. This shows that the algorithm of the present invention has achieved good results in the extraction and processing of detailed information.

Description of drawings

Fig. 1 is a flow chart of an underwater image enhancement method based on weighted fusion;

Figure 2 is a comparison effect diagram of the results of different algorithms;

Fig. 3 is a schematic diagram of underwater image mean value comparison;

Fig. 4 is a schematic diagram of average gradient comparison of underwater images;

Fig. 5 is a schematic diagram of underwater image information entropy comparison;

Fig. 6 is a schematic diagram of standard deviation comparison of underwater images.

detailed description

In order to make the purpose, technical solution and advantages of the present invention clearer, the implementation manners of the present invention will be further described in detail below.

Due to the attenuation and scattering of light during underwater transmission, underwater images are severely degraded, resulting in loss of detail, loss of contrast, and color distortion. Although the existing underwater image enhancement methods have made progress to a certain extent, the image processing effect is still not satisfactory. In addition, the related algorithms are complex, or based on a specific condition, and the application range is limited.

Example 1

In order to improve the clarity of underwater images, the embodiment of the present invention proposes an underwater image enhancement algorithm based on weighted fusion (Enhancing Underwater Images based on Weighted Fusion, EUIWF). Compared with other existing enhancement algorithms, this algorithm is Simple pixel operation, concise and clear, does not require complicated deconvolution operations, and reasonable selection of weights makes the image have richer details on the basis of color correction.

101: The degraded underwater image is processed by Gray-World (grayscale world) and histogram equalization respectively to obtain the input image;

102: Through normalization processing, the method of weighted guided filtering is used to realize the definition of the weight factor, and the weight factor is corrected to obtain the corrected weight map;

103: Use the Laplacian pyramid to decompose each input image separately, and use the Gaussian pyramid to decompose each weight image, and finally use the multi-scale fusion method to fuse the input image and the weight image to obtain an image with rich details .

In summary, the embodiment of the present invention combines the weight factor and the Laplacian pyramid fusion through the above steps 101 to 103, without the need for complex deconvolution operations, and reasonably selects the weights, so that the image Richer detail on top of color correction.

Example 2

The scheme in embodiment 1 is further introduced below in conjunction with specific calculation formulas and accompanying drawings, see the following description for details:

The embodiment of the present invention fuses the input image and the weight map based on the fusion method of Ancuti. In the basic fusion method, the weight plays an important role, and it has a decisive effect on the clarity, saliency and contrast of the image. Therefore, the results obtained by simply normalizing the four weight maps and directly adding them cannot Weighing which of the four features of the pixel is more dominant, the feature information of the image cannot be reasonably allocated, so that the image loses some details. In order to obtain an image with rich details, the embodiment of the present invention corrects the weight map as follows:

m_W ₁ ＝a_1*n_W _LC 1+b_1*n_W _S 1+c_1*n_W _E 1 (7)

m_W ₂ ＝a_2*n_W _LC 2+b_2*n_W _S 2+c_2*n_W _E 2 (8)

Among them, a_i, b_i, and c_i are the scaling factors of each weight, respectively, and n_W _LC i, n_W _S i, and n_W _E i are the normalized weight results of the local contrast weight, saliency weight, and exposure weight, respectively.

(n_W _LC i=W _LC i/(W _LC 1+W _LC 2), similarly to the other two weight graphs n_W _S i and n_W _E i, i=1, 2) The significance of this weight factor is that when it is at the boundary , the influence parameter is greater than 1, and the weight is in a dominant position. At this time, it is easier to highlight the boundary, improve the definition, and enhance the edge during fusion.

201: Use Gray-World and histogram equalization processing on degraded underwater images respectively to obtain input images I ¹ and I ² ;

202: Calculate the local contrast weight, saliency weight, and exposure weight of each input image respectively, perform normalization processing, and correct its weight factor to obtain a corrected weight map;

In order to enhance the edge and remove the interference of noise, the embodiment of the present invention adopts the method of weighted guided filtering to realize the definition of the weight factor, that is:

Among them, ε is a very small positive number to prevent the denominator from being zero.

203: Use the Laplacian pyramid to decompose each input image separately, and use the Gaussian pyramid to decompose each weight image, and finally use the multi-scale fusion method to fuse the input image and the weight image to obtain an image with rich details .

In summary, the embodiment of the present invention combines the weight factor and the Laplacian pyramid fusion through the above steps 201 to 203, without the need for complicated deconvolution operations, and reasonably selects the weights to make the image Richer detail on top of color correction.

Example 3

The scheme in embodiment 1 and 2 is carried out feasibility verification below in conjunction with specific example, see the following description for details:

In order to verify the effect of the algorithm, the enhanced image is processed by applying the above-mentioned algorithm to the underwater image. The underwater images are all color images, so the three channels of R, G, and B are calculated separately during fusion. In order to compare the image quality of the weighted fusion algorithm, the embodiment of the present invention will be compared with the existing algorithms, for example, the underwater enhancement algorithm based on dehazing processing [2], the enhancement method based on Retinex [6], and the basic fusion Algorithm [7].

The objective experiment uses image mean, average gradient, information entropy and standard deviation four image quality measurement indexes to measure the experimental results. The average value of the image reflects the average brightness of the image; the average gradient is the clarity of the image, which reflects the ability of the image to express the contrast of details; the information entropy reflects the amount of information contained in the image, and is an important indicator to measure the richness of image information; standard The difference reflects the degree of dispersion of the gray mean.

It can be seen from Figure 2 that the algorithm of Fattal et al. has a good effect in defogging, but it is not very effective in color correction, because its method requires sufficient color information, and the amount of information in the G and B channels of underwater images is relatively large. Large, so the color compensation for the R channel is lost. The algorithm of Xueyang Fu et al. is overall dark and lacks detailed information. Similarly, the algorithm of Ancuti et al. does not distribute the weights proportionally, so some areas are dark and the details are not obvious.

Adopting the technical solution of the present invention sees that the overall sensitive effect is better from a subjective point of view, and in terms of objective indicators, Fig. 3 to Fig. 6 have shown that this method is used for five images in Fig. 2 on image mean, average gradient, information entropy and standard deviation From the results, the results of this method are higher than the other three methods. This shows that this method is superior to the other two algorithms in the extraction and processing of detailed information.

In practical applications, in order to obtain the best enhancement results, the parameters involved in this method are set as follows: σ=0.25 in formula (3), ε=10 ^-4 in formulas (9), (10) and (11) . Taking the degraded underwater image as the experimental object, using the steps described in this method and setting the above parameter values, a fused image with better visual effect can be obtained. Experimental results show that this method has good results in both subjective vision and objective quantitative indicators.

Those skilled in the art can understand that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the above-mentioned embodiments of the present invention are for description only, and do not represent the advantages and disadvantages of the embodiments.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (3)

1. a kind of underwater picture Enhancement Method based on Weighted Fusion, it is characterised in that the underwater picture Enhancement Method includes Following steps：

Gray-World and histogram equalization processing are used respectively to the underwater picture degraded, input picture is obtained；

By normalized, the definition to weight factor is realized using the method for weighting guiding filtering, weight factor is corrected, obtains To revised weight map；

Each width input figure is decomposed respectively with laplacian pyramid, and to each width weight map gaussian pyramid Decompose, the final method with Multiscale Fusion is merged to input picture and weight map, obtain the abundant image of details.

2. a kind of underwater picture Enhancement Method based on Weighted Fusion according to claim 1, it is characterised in that described The step of to revised weight map is specially：

m_W₁=a_1*n_W_LC1+b_1*n_W_S1+c_1*n_W_E1

m_W₂=a_2*n_W_LC2+b_2*n_W_S2+c_2*n_W_E2

Wherein, a_i, b_i and c_i are the scale factor of each weight, n_W respectively_LCI, n_W_SI and n_W_EI is respectively above-mentioned office The normalized weight result of portion's contrast weight, conspicuousness weight and exposure weight；I=1,2.

3. a kind of underwater picture Enhancement Method based on Weighted Fusion according to claim 1, it is characterised in that described to adopt The step of being realized with the method for weighting guiding filtering to the definition of weight factor be specially：

$a\_i=\frac{n\_W_{LC}i}{\frac{1}{N}({\mathrm{\Σ}}_{k=1}^{K}n\_W_{LC}k+\mathrm{\ϵ})}$

$b\_i=\frac{n\_W_{s}i}{\frac{1}{N}({\mathrm{\Σ}}_{k=1}^{K}n\_W_{s}k+\mathrm{\ϵ})}$

$c\_i=\frac{n\_W_{E}i}{\frac{1}{N}({\mathrm{\Σ}}_{k=1}^{K}n\_W_{E}k+\mathrm{\ϵ})}$

Wherein, ε is the positive number of a very little, and it is zero to prevent denominator；n_W_LCK, n_W_SK and n_W_EK is respectively local contrast power The normalized weight of weight, conspicuousness weight and exposure weight；K is 2；N is sum of all pixels.