Since the advent of the semiconductor laser, the rapid development, many achievements, and wide applications make it almost occupy "half the sky" of the laser market. As an emerging laser light source, the white laser is an illumination light source technology that generates high-brightness white light through the laser^[1-2]. The white laser has the advantages of wide frequency coverage, high brightness, high power peak value, strong directivity, a long working life, strong spatial and temporal coherence, etc. Therefore, it has played a practical role in scientific research and national defense, lighting display, communication information, medical testing, industrial production, and other fields^[3-4].

With the increasing demand to detect and understand the ocean, the development of underwater laser imaging, underwater spectral analysis, underwater lighting, and other fields has grown rapidly^[5-7]. At present, the light sources commonly used underwater imaging include LEDs, tungsten halogen lamps, and lasers. LED white light sources are widely used in underwater imaging and underwater target recognition due to their advantages of low cost, high luminous efficiency and long service life, but their energy is more divergent than laser light sources and their underwater transmission distance is shorter, so they are not suitable for underwater long-distance imaging and target recognition. Since seawater has much less blue-green light attenuation with wavelengths in the 470−580 nm band than other light bands, many countries have invested a lot of manpower and material resources in the application of blue-green lasers for underwater communication, detection, sensing, and other fields. Although blue-green lasers play an important role in applications such as underwater communication and detection, due to the limited spectral range of monochromatic lasers, blue-green lasers cannot be well used in underwater target recognition and spectral analysis^[8-9]. The emergence of the semiconductor white laser provides a new development direction for the research of underwater light sources. Semiconductor white lasers can not only cover the visible light spectrum, but also have relatively concentrated energy. Therefore, the application of a white laser in underwater imaging can not only overcome the narrow spectral range of monochromatic lasers, but also overcome the LED white light source energy divergence and short transmission distance. This is of great significance to the development of deep-sea detection technology and underwater laser imaging, and can better meet the application requirements of target imaging and identification for deep-sea detection instruments.

At present, due to limitations to the development of semiconductor lasers and the lack of in-depth research on synthetic white lasers, researches about white laser light sources used for underwater detection has not been carried out.

At the same time, due to the complexity of the water environment and the scattering and absorption of light in water, the underwater imaging quality is poor, which seriously limits the further development of underwater detection technology^[10]. Underwater image processing technology can help improve image quality and effectively extract target information, which not only provides technical support for improving the performance of underwater target detection systems, but also provides reference directions for the image analysis of underwater white laser imaging.

In this paper, an underwater laser imaging system is built with a white laser used as the underwater lighting source with a power of 220 mW and a color temperature of 6469 K based on the red, green, and blue (RGB) three-primary laser synthesis, then a series of experiments are carried out. The purpose of the experiment is to explore whether the white laser can meet the working requirements of underwater laser imaging. Then, the underwater imaging effects of the white laser, the red, green, and blue monochromatic lasers, and the LED white light sources were compared under different underwater environments, different underwater imaging distances, and different targets. On this basis, we discuss whether three traditional image processing methods, namely median filtering, histogram equalization, and piecewise linear grayscale transformation, are suitable for underwater white laser imaging. At the same time, through the image processing and quality evaluation of underwater LED white light sources and white laser imaging, the superiority of underwater white laser imaging is verified.

The experimental principle of using a camera to shoot underwater objects under a specific light source is shown in Figure 1. The underwater light source is fixed at the upper left corner of the water tank to ensure that all the light of the light source can be injected into the water, and the target is placed in the center of the water tank’s bottom so that it can move in vertically. A ruler is used to measure the distance of the target from the water surface. The camera is positioned on the water and directly above the object to capture the image.

The experimental instrument used in this paper is the LWRGB-F-FTP laser (LaserWave, China). It has the advantages of stable power, high beam quality, and convenient operation. It composed of power supply, laser head, optical fiber and lens barrel. The white laser set up synthesized by an RGB semiconductor is shown in Figure 2.

In this experiment, a 638-nm semiconductor red laser, 520-nm semiconductor green laser, and 445-nm semiconductor blue laser are used. After fiber coupling and output, the fiber core diameter is 400 μm. After the fiber is subjected to flat-top homogenization, a flat-top spot is obtained. The uniformity is larger than 70% and the angle is 29°. The lens barrel is connected to the optical fiber light outlet, and the light is finally emitted through the lens barrel. The optical parameters of the output white laser are tested by a spectral illuminometer (OHSP-350, Hopoocolor, China), and an optical power meter (LI-P50W, LaserWave, China) is used to measure the laser output power.

The white laser used in this paper is synthesized by three primary color lasers. The color temperature of the white laser is 6469 K, the color coordinates are (0.3199, 0.2805), and the power is 220 mW. The output spectrum of the white laser is shown in Figure 3 (color online).

The monochromatic lasers used in this paper includes red, green and blue primary color lasers. The color coordinates of the red laser are (0.7165, 0.2835) and the power is 260 mW; the color coordinates of the green laser are (0.0554, 0.8247) and the power is 120 mW; the coordinates of the blue laser are (0.1598, 0.0149) and the power is 330 mW. The output spectrum of the three primary color lasers are shown in Figures 4, 5 and 6.

The peak wavelengths corresponding to the three primary color lasers are 637.4 nm for the red light, 517.4 nm for the green light, and 446.6 nm for the blue light, respectively. Their spectral half-widths are 4.2, 4.5, and 4.0 nm, and the corresponding irradiances are 23.34, 17.45, 29.03 mW/cm².

The relevant optical parameters of the LED white light source used in this paper are shown in Table 1.

The shooting device used in this paper is a Huawei Honor P30 mobile phone, and the rear camera is used to shoot. The rear camera consists of three lenses: a 40-megapixel photosensitive lens, a 16-megapixel wide-angle lens, and an 8-megapixel telephoto lens.

This paper mainly discusses the median filter method, histogram equalization, piecewise linear grayscale transformation, the MSRCR enhancement algorithm, and the Laplacian pyramid fusion method.

This method can effectively reduce noise and eliminate abnormalities, especially for removing impulse noise interference. The principle is to establish a window, and the odd-numbered pixel values in the window are arranged in order of size, such as 3×3, 5×5, etc., and the median value of the window is taken to replace the gray value of the center point. In this paper, we will use 3×3, 5×5, 7×7, and 9×9 templates to perform median filtering on the collected images respectively.

The principle is to select the grayscale histogram of the original image and convert the more concentrated grayscale interval into a uniform distribution of all the grayscale intervals of the histogram. After the histogram is equalized, the grayscale range of the image is expanded, so the contrast is enhanced.

The piecewise linear grayscale transformation algorithm maps the original image grayscale range 0~P to the transformed image grayscale range 0~Q through the transformation function and sets the original image function as f(x, y), and the transformed image function as g(x, y). If a given gray level range of the original image is [a₀, b₀], and the given gray level range of the transformed image is [a₁, b₁], then the transformation function is:

$ g(x,y) = \left\{ {\begin{aligned} &{({a_1}/{a_0})f(x,y),0 \leqslant f(x,y) < {a_0}} \\ &[({b_1} - {a_1})/({b_0} - {a_0})][f(x,y) - {a_0}] + {a_1}, \\ &{a_0} \leqslant f(x,y) \leqslant {b_0} \\ &[(Q - {b_1})/(P - {b_0})][f(x,y) - {b_0}] + {b_1}, \\ &{b_0} < f(x,y) \leqslant P \\ \end{aligned}} \right. . $ (1)

When the values of a₀, b₀, a₁, b₁ change, the corresponding image processing results are also different. That is, when a₀>a₁, b₀<b₁, the original image gray level is expanded in the range of [a₀, b₀], and [0, a₀] and [b₀, P] is compressed; when a₀<a₁, b₀>b₁, the gray level of the original image is expanded in the range of [0, a₀] and [b₀, P], and [a₀, b₀] range is compressed. The image contrast can be increased through the above algorithm.

The algorithm formula is as follow:

$ r(x,y) = \log R(x,y) = \log \frac{{S(x,y)}}{{L(x,y)}} \quad, $ (2)

where R(x, y) represents the reflection property, L(x, y) and S(x, y) represent the incident light and the original image.

The application formula is expressed as follow:

$ r(x,y) = \log S(x,y) - \log [F(x,y) \oplus S(x,y)]\quad . $ (3)

Among them, r(x, y) is the output image, $\oplus $ represents the convolution operation, and F(x, y) is the center wrapping function.

$ F(x,y) = \lambda \cdot {{\rm{e}}^{ - \tfrac{{{x^2} + {y^2}}}{{{c^2}}}}} \quad, $ (4)

where c is the Gaussian surround scale and λ is a constant.

The output images are weighted and averaged:

$ r(x,y) = \sum\limits_k^K {w_k}\log S(x,y) - \log [F(x,y) \cdot S(x,y)] \quad . $ (5)

Among them, K is the number of Gaussian centers, and multiple w_k are evenly distributed.

The output image is added with a color recovery factor C(x, y):

$ C(x,y) = f[{I'_i}(x,y)] = \beta \log [\alpha {I'_i}(x,y)]\quad, $ (6)

where f(·) is the mapping function, and α and β represent gain constants and nonlinear strengths.

The high-frequency and low-frequency signals are obtained by decomposing the image, and inversely transformation of the two signals are carried out to reconstruct the image, and finally, the image is obtained by fusion. The fusion formula is as follow:

$ {R^{l}}(x,y) = \sum\limits_{k = 1}^K {{G^l}\{ {W^k}(x,y)\} {L^l}\{ {I^k}(x,y)\} }\quad,$ (7)

among them, l is the pyramid level, L{I} is the Laplace transform of the input image I, and G{W} is the normalized weight of the Gaussian distribution.

The quality evaluation indicators used in this paper are Peak Signal-to-Noise Ratio (PSNR) and image structural integrity.

PSNR reflects the completeness of image detail information. The larger the value, the more complete the image information. The formula is as follow:

$ PS NR = 20 \cdot \lg\left(\frac{{MA{X_l}}}{{{\kern 1pt} \sqrt {MS E} }}\right) \quad,$ (8)

where MSE is the mean square error between the original image and the processed image, and MAX is generally 255.

Structural Similarity Image Integerity (SSIM) for short, the larger the value, the better the structural integrity of the image. The formula is as follow:

$ SSIM(x,y) = \frac{{(2{u_x}{u_y} + {C_1})(2{v_x}{v_y} + {C_2})}}{{({u_x}^2 + {u_y}^2 + {C_1})({v_x}^2 + {v_y}^2 + {C_2})}} \quad, $ (9)

where u_x and u_y are the image’s mean values, v_x and v_y are the image’s standard deviations, and C₁ and C₂ are constants.

The experiment was carried out in a dark field environment, and the ambient light intensity was less than 1lx. The experimental conditions are as follows: (1) underwater light sources are white laser; red laser; green laser; blue laser. (2) water quality: seawater, clear water; (3) underwater imaging distance: 5 cm, 19 cm; (4) underwater targets: oranges, safflowers, green leaves; (5) sink: 42 cm ×23 cm × 26 cm transparent glass sink.

The laser is fixed to the upper left corner of the water tank through the bracket, and the target is placed in the water tank, and it is in the center of the camera's field of view. Under varying conditions of seawater, clear water, and no water and the distance between the target and the water surface, a white laser and three monochromatic red, green, and blue lasers are used for contrast illumination respectively. The camera are adjusted to shoot objects at 5 cm and 19 cm underwater, and compare the lighting effects of the four light sources. The experimental process is shown in Figure 7 (color online).

When the water tank contains clear water, the images of different objects under different light sources and different imaging distances are collected, as shown in Figures 8-9 (color online).

When the water tank contains seawater, the images of different objects under different light sources and different imaging distances are collected, as shown in Figures 10-11 (color online).

An image of the safflower at a distance of 19 cm from the sea water surface illuminated by the white laser in the above experiments was processed, and the applicability of the image smoothing processing based on median filtering and image enhancement using histogram equalization and piecewise linear grayscale transformation were discussed. The processing results are shown in Figures 12-14 (color online).

Based on the analysis of the images of different objects illuminated by different light sources at different underwater imaging distances in clear water and seawater, we can derive the following conclusions:

(1) Under the same underwater imaging distance and camera parameters, the color reproduction of the images illuminated by the white laser light source is higher than that of the images illuminated by the red, green, and blue lasers.

(2) When the imaging distance is long, the color of the object illuminated by the white laser is still clearly visible, while the color and edge information of the object under other light sources are difficult to identify.

(3) Due to a large number of particles and impurities in seawater and the complex water environment, the quality of the images obtained in seawater is not as good as the images in clear water.

(4) These three kinds of image processing methods can improve the image contrast to a certain extent, but as the underwater imaging distance increases, the image edge information will also be blurred.

The experimental results show that the illumination effect of the white laser for underwater laser imaging is better than that of the red, green, and blue monochromatic lasers.

In the underwater laser imaging system built in Section 3, image acquisition is performed after changing certain experimental conditions. The transformed experimental conditions are as follows:

(1) Underwater light source: white laser with an output power of 220 mW and color temperature of 6469 K; LED1 with an output power of 234 mW and color temperature of 7261 K; LED2 with an output power of 242 mW and color temperature of 5846 K;

(2) Water quality: seawater, clear water;

(3) Underwater imaging distance: 15 cm, near the water surface;

(4) Underwater target: target A- small wheel, target B- breakfast milk;

(5) Sink: 42 cm × 23 cm × 26 cm transparent glass sink.

The laser is fixed to the upper left corner of the water tank through the bracket, and the target is placed in the water tank and is in the center of the camera's field of view. The distance between the target and the water surface are changed with a white laser and two LED white light sources for contrast illumination. Adjust the parameters of the camera to be consistent, shoot underwater objects at 15 cm and near the water surface, and compare the lighting effects of the light source. The experimental process is shown in Figures 15 and 16 (color online).

In Figures 15-16, the images obtained with different light sources LED1, LED2, and white laser in seawater are processed through histogram equalization enhancement AHE, Laplace pyramid fusion, and the MSRCR color restoration algorithm. The processing results are shown in Figures 17-19 (color online).

After analyzing the above processing results, it is subjectively believed that in terms of processing effect, Laplacian Pyramid Fusion is better than Histogram Equalization AHE and Histogram Equalization AHE is better than MSRCR Color Restoration Algorithm.

We recorded the image quality parameters after the Laplacian pyramid processing at this time, and on this basis, for the image processed by the Laplacian pyramid, image enhancement is performed by the contrast limited adaptive histogram equalization method (“CLAHE” for short). The image quality evaluation was implemented for the processing results of each step, and the underwater imaging performance of the white laser and the LED white light source was compared and analyzed. The flowchart of the image processing and evaluation method is shown in Figure 20.

The processed results are shown in Figures 21 and 22 (color online).

The quality of above processed images are evaluated, and the quality evaluation indicators are PSNR and SSIM. The results are shown in Tables 2 and 3.

By analyzing the images, we can conclude:

(1) Comparing the images under different light sources and different water qualities, it can be found that the seawater has more particles and impurities and the water environment is more complex, resulting in the quality of the taken images is not good as the images in the clean water environment. When the LED white light is imaged in the underwater environment, there is a sudden drop in optical power. The propagation path of the LED white light is more divergent than that of the white laser, the attenuation effect of the water body is stronger, and the light scattering reaction is more serious. As a result, the image quality of white light in underwater environment becomes worse. The utilization efficiency is low, the image presented is relatively blurred and brightly exposed, and some of the characteristic information of the target is lost.

(2) By comparing image qualities under different light sources and different imaging distances, it can be found that when the imaging distance is long, the feature information of the target under white laser illumination is still clearly visible, while the color and edge information of the target under an LED white light source and other features have been difficult to identify.

To summarize, the illumination effect of a white laser light source for underwater laser imaging is better than that of an LED white light source.

By analyzing the PSNR and SSIM values ​​in Table 2 and Table 3, we can conclude:

(1) By comparing the PSNR and SSIM values under the white laser and the white LED light source, it can be found that the value under the white laser is larger, indicating that the white laser is better than the LED white light in terms of retaining image details and image structure integrity;

(2) By comparing the values of objects A and B under the two light sources, it can be found that there are differences in the values, and the difference under the white laser light source is smaller than that under the LED white light source, indicating that the laser light source has a stronger ability to store target information and the imaging robustness under the laser light source is stronger;

(3) By comparing the performances of the four image processing methods for underwater images under white laser, it is found that the PSNR value of the image processed by CLAHE is the largest, indicating that this method maximizes the feature and detail information of the image to the maxmum extent. The SSIM value of the image processed by the Laplacian pyramid fusion method is the largest, indicating that this method is good at preserving image features and highlighting edge information.

From the above analysis results, it can be concluded that the white laser light source is more suitable for underwater imaging than the LED light source.

According to the principle of underwater laser imaging, the white laser light source synthesized by red, green, and blue semiconductor lasers is used as an underwater light source to realize the image acquisition of underwater imaging, which provides a new direction for the research of underwater lighting sources. With red, green, and blue monochromatic lasers and LED white light sources as the contrast light source, the underwater lighting effect test experiment was carried out. The experimental results show that the white laser has high color reproduction, a long imaging distance, and richer target image information under underwater imaging condition. It proves the superiority of the white laser as an underwater illumination source. In addition, when the white laser beam combined with the semiconductor three-primary laser beams is used as the light source of the underwater detection system, it can not only be used as a white light source, but also can be easily realized by controlling the output of three single-color lasers and using only a single-color laser for illumination. The mutual switching between the color light source and the white light source is conducive to promoting the multi-functional compatibility and performance expansion of the underwater detection system, and is of great significance to the development of deep-sea detection technology. The combined semiconductor beam and white laser light source has both the advantages of monochromatic laser transmission distance and the spectral advantages of LED white light sources, and is a new research direction of underwater lighting.