Sea surface clutter suppression method based on time-domain polarization characteristics of sun glint

Jian-An Liang; Xia Wang; Si He; Wei-Qi Jin

doi:10.1364/OE.27.002142

1. Introduction

Because a three-dimensional (3-D) image signal (comprising a 2-D space signal and a 1-D time signal) can provide more abundant information, search and tracking systems based on 3-D image sequences have proliferated. Among these, infrared search and tracking systems (IRSTs) are considered to be among the most promising for use in military applications owing to their all-weather capabilities and strong anti-interference characteristics. However, target detection at a low signal-to-clutter ratio (SCR), particularly in the presence of sun glint, has always been a problem in the application of infrared imaging technology.

Polarization imaging technology can effectively suppress sea clutter. In an infrared polarization imaging system, the polarizer’s transmission direction is set perpendicular to the main polarization direction of the sun glint [1,2], which in the mid-wave infrared band is primarily horizontal with a widely varying degree of polarization [3,4]. However, if the sun glint is very strong, infrared polarization imaging can only suppress some of the clutter, and the residual clutter can still seriously impair the detection of a target. Zhao et al. used an infrared polarization imaging system with two angle adjustable polarizers to further suppress sun glint. Although their technique achieved good results, it was complicated to implement [5]. Previously, we proposed a clutter suppression method based on the co-application of an image processing algorithm with polarization imaging technology [6]. Using the strong intensity of the sum of the clutter polarization components, our method applied the RX anomaly detection algorithm to select the clutter, with the image filtering algorithm used to suppress the residual clutter. Experimental results demonstrated that the method was effective in suppressing sea clutter.

Current applications of infrared polarization imaging technology for suppressing clutter are overwhelmingly based on the use of 2-D spatial signals and rarely involve time signals. IRSTs generally use clutter suppression algorithms based on spatial, temporal, or spatiotemporal approaches to suppress background clutter. Because 2-D spatial clutter suppression algorithms can only consider current frame data, they are not skillful at suppressing complex backgrounds [7]. 1-D time domain clutter suppression methods rely on the strong correlation within the background region in the time domain to use the previous frame image to predict the background of the current image and suppress the clutter based on the difference image, effectively improving detection probability [8,9]. Clutter suppression methods based on the 3-D spatiotemporal domain fully combine the advantages of spatial filtering and temporal filtering to significantly improve tracking accuracy and effectiveness [10,11]. Kim et al. proposed a separate spatiotemporal filtering method based on image attribute correlation that effectively suppresses the bulk of cloud and sea clutter while reducing the false alarm rate in small target detection [12]. Although improvements in the clutter suppression effect are often accompanied by an increase in data dimension and algorithm complexity, which makes algorithmic application in real-time or at scale difficult, the current push toward more efficient clutter suppression has generally involved increasing the data dimension, changing the imaging mode, and/or enhancing the performance of the algorithm. In our previous work, we demonstrated that combining polarization imaging technology with a clutter suppression algorithm can suppress sea clutter [6]. Following these results, we have continued to study the polarization characteristics of sea surface clutter in the time domain with the goal of combining polarization imaging technology with a 1-D time domain-based data processing algorithm to obtain a more efficient clutter suppression method.

In this paper, the concept of clutter suppression based on polarization information is extended to the time domain with the goal of presenting a more effective clutter suppression method. The remainder of this paper is organized as follows. In Section 2, the concept of time-domain polarization characteristics is proposed and the time-domain polarization characteristics of sun glint are analyzed. The proposed method is outlined in Section 3, while Section 4 discusses our experimental results and conclusions. Discussion about the method is in Section 5.Finally, the paper is summarized in Section 6.

2. Simulation study on time-domain polarization characteristics of sun glint

Sun glint is formed by specular reflection from surface tension waves produced by wind [13,14]. A typical water surface will not be static but will instead fluctuate and, under the sun's radiation, form numerous flares that appear and disappear quickly [15]. The sun glint formed by reflected light in the mid-wave infrared band comprises partially polarized light with a degree of polarization related to the wavefront distribution. Wang et al. found that the polarization characteristics of the sun glint entering a receiver’s field of view (FOV) are altered by changes in the wave surface [16], and related studies have shown that the sea surface as a whole can be seen as a collection of high and low fluctuating micro-wavelet surfaces with varying slopes and directions [17,18]. The sun glint in each micro-plane can be considered to be the result of specular reflection with polarization characteristics varying with changes in the slope of the micro-plane [19].

The amount, appearance, and disappearance of sun glint are related to changes over time in the sea surface wavefront; more technically, the characteristics of sun glint can be said to vary in the time dimension. As the polarization characteristics of sun glint in the infrared band are also related to the time-varying wavefront, we can also hypothesize that sun glint has a time-domain polarization characteristic. To address the lack of research in the literature on the temporal polarization characteristics of sun glint, we investigated the time-domain polarization characteristics of sun glint in the mid-wave infrared band with the goal of using these characteristics to develop a more effective sea clutter suppression method.

2.1 Dynamic sea surface infrared polarization radiation model

Although the radiation characteristics of a rough sea surface can be assessed through both theoretical modeling and experimental observation, the significant investment required to carry out experimental observation and limitations in obtainable experimental conditions have made the theoretical approach increasingly popular among researchers. We previously established a sea surface infrared polarization radiation model [20] based on a two-scale method that uses Elfouhaily spectra to generate random waves on a simulated sea surface under different wind speeds with a modeled slope, i.e., wave height distribution, conforming to existing empirical formulation. Using this approach, an infrared polarization radiation model of the sea surface was established based on a ray reverse tracking method that applied Monte-Carlo and binary tree acceleration algorithms. Finally, a ray tracing method was used to accurately simulate the infrared polarization radiation scene of the sea surface. This method, however, did not consider the time factor; in the following subsection, the model is augmented by the inclusion of a time factor to better capture the dynamic polarization characteristics of the sea surface.

Our previous approach used a wave model based on statistical properties. Typically, the wave height field will be augmented by a superimposed set of sine or cosine wave components with different frequency and amplitudes. Technically, the mathematical process underlying this involves the application of an inverse Fourier transform. Regarding the height of a wave H (x, t) as a function of position x and time t, the wave height field can be expressed in terms of multiple complex amplitudes and time-dependent sinusoidal functions:

H (x, t) = \sum_{k} \tilde{H} (k, t) \exp (i k \cdot x),

where x = (x, z) is the sea surface coordinate position,

k = (k_{x}, k_{z})

is the 2-D vector of the wave number, and

\tilde{H}

(k, t) represents the amplitude corresponding to the Fourier series when the wave number is k and the time is t. The value of

\tilde{H}

(k, t) depends on the distribution of the wave spectrum at different wind speeds:

\tilde{H} (k, t) = {\tilde{H}}_{0} (k) \exp [i ω (k) t] + {\tilde{H}}_{0}^{*} (- k) \exp [- i ω (k) t]

{\tilde{H}}_{0} (k) = \frac{1}{\sqrt{2}} (ξ_{r} + i ξ_{i}) \sqrt{Ψ_{h} (k) Δ k_{x} Δ k_{z}},

where ${\tilde{H}}^{*} (k, t)$ is the conjugate function of $\tilde{H} (k, t)$ , ξ_i and ξ_r are independent Gaussian random numbers with mean values of zero and standard deviations of one (other random numbers, such as lognormal random numbers, can also be used), Ψ_h( $k$ ) is the sea surface wind direction spectrum (in our previous work [20], the Elfouhaily spectrum was used), Δk_x and Δk_z are the unit intervals of the discrete wave numbers in the x and z directions, respectively, and ω(k) is the angular frequency corresponding to the wave number, with ω(k) and k satisfying a specific dispersion relation. Because the sea surface can be considered a combination of multiple different amplitudes, phases, and wave numbers, the entire waveform will change over time if the phase velocity is a function of the wave number. In the deep-sea region, the influence of the seabed can be ignored and the dispersion relationship can be expressed as [21]

ω^{2} (k) = g k,

where g = 9.8m/s² is the gravitational acceleration.

The infrared polarization radiation model of the sea surface described in [19] can be augmented by adding a time factor and using Eq. (4) to obtain the wave spectrum distribution at different times using Eq. (2). In turn, Eq. (1) can be used to obtain the time-varying wave height field from which the polarization radiation model of the dynamic sea surface can be obtained. We applied this model to analyze the polarization radiation characteristics of a simulated dynamic sea surface.

2.2 Simulation results and analysis

The simulations were conducted using the Modtran mid-latitude summer atmospheric model with a sea water temperature of 293K, a sea surface wind speed of 2 m/s, and a solar zenith angle of 80°. The observation radius and nadir angle were 50 m and 70°, respectively. The detector was a mid-wave infrared detector with a resolution of 320 × 240, focal length of 50 mm, and a pixel pitch of 30 μm. Analysis occurred over an 8 m × 8 m surface area and involved two scenes. In the first, the polarization radiation data of the sea surface were collected every 0.05s over a total acquisition duration of 3 s, close to the frame rate of the video captured by a typical camera. In the second scene, the polarization radiation data were collected every 3 s over a total acquisition duration of 108 s, a setting compatible with the experimental setup described in the next section. Using the model, the Stokes vector, linear degree of polarization (DOLP), and polarization angle (AOP) of the real-time sea surface were obtained.

The simulation results at three instants (t = 0, 0.05, and 0.1 s) of scene one and three instants (t = 3, 6, and 9 s) of scene two are selected as the simple illustration, with intensity (I), DOLP, and AOP images obtained by the simulation shown in Fig. 1. It is seen that each I image has pixels that are much brighter than the typical pixel brightness. These pixels mark the positions of sun glint and, over time, the position and number of sun glints are seen to change. The DOLP images also contain pixels with high DOLP values, the positions of which correspond to the sun glint pixels in the corresponding I images. The sea surface AOP also changes over time to reflect the time-varying infrared polarization characteristics of the sea surface and, therefore, sun glint.

Fig. 1 I, DOLP, and AOP simulation images at three different instants in (a) scene one and (b) scene two.

Download Full Size | PDF

We next assessed the polarization characteristics of the sun glint. To do this, we separately counted the changes in the number of sun glints with time in the I and DOLP images of scenes one and two, as shown in Fig. 2. It was determined that, for both scenes, the position and number of sun glints at various times in the I and DOLP images were statistically consistent, i.e., the I- and DOLP-based statistical curves in Fig. 2 are coincident. This demonstrates that the sun glint in the mid-wave band has a strong polarization characteristic under the chosen simulation setting.

Fig. 2 Number of sun glints appearing in scenes one (left) and two (right) as functions of time.

Download Full Size | PDF

Although the time variances in the number of sun glints in Fig. 2 follow no apparently regular pattern, within a specific time window the variance will always fall within a certain range: the range of scene one is [30, 55], while that of scene two is [28, 56]. Thus, although the number of sun glints varies with time in the dynamic sea scene, it does so within a constrained range of fluctuation under constant imaging conditions which means the imaging environment and detection equipment remain unchanged.

We then counted the number of the sun glints appearing in each pixel position over the total durations of scenes one and two, respectively. Figure 3 illustrates the statistical results, with the occurrence frequencies encoded as pixel colors, while Table 1 lists the actual numerical results, including the number of effective pixels involved in the statistics, the occurrence number of sun glint pixels, and the ratio of the latter to the former.

Fig. 3 Statistics on the number of sun glints occurring over time at each pixel position in scenes one (left) and two (right) over the modeling period. The occurrence frequencies of each pixel are color coded.

Download Full Size | PDF

Table 1. Statistics on sun glint recurrence at pixel positions in scenes one and two over the modeling period

View Table | View all tables in this article

It is seen from Fig. 3 that the positions of sun glint over the modeled interval are widely distributed over the sea surface area. The pattern reflects the time-varying positions of the sun glints and the range within which the number of glints is constrained. Although the sun glint within a pixel can remain fixed for a period of time, it can also appear at and disappear from any pixel. An examination of Table 1 reveals that there can be up to four and three glint recurrences within a given pixel in scenes one and two, respectively, which confirms the ability of the sun glint to recur multiple times at a given position, although with a frequency that differs by statistical time period length and interval. It is also seen that most pixels experiencing a sun glint will do so only once. In scene one, there are 2,223 pixel positions with single sun glint occurrences, accounting for 9% of the effective pixels and 93.4% of the total number of pixels in which sun glint appears. In scene two, there are 1,448 pixel positions with single sun glint occurrence, accounting for 5.9 and 95.3% of the effective and total number of pixels in which sun glint appears, respectively. It can thus be concluded that, although sun glint can return to a position multiple times, most positions experiencing sun glint will do so only once. In addition, the position and frequency of sun glint will vary with the time statistics used to analyze the scene.

We then compared the changes in the I and DOLP values of the sun glint and non-glint regions over time, with the results shown in Fig. 4, which shows the mean values of radiation in the respective regions (the intensity mean value is normalized to [0, 1]). The ranges of the I and DOLP mean values are listed in Table 2.

Fig. 4 I and DOLP mean values of sun glint and non-glint regions in (a) scene one and (b) scene two over time.

Download Full Size | PDF

Table 2. Ranges of intensity (I) and DOLP mean values of sun glint and non-glint regions in scenes one and two changing over time

View Table | View all tables in this article

It is seen from Fig. 4 that the I and DOLP mean values in both the sun glint and non-glint regions fluctuate within specific ranges over time, with the range of fluctuation of the former much higher than that of the latter. Furthermore, the mean values are much higher within the sun glint regions. An examination of Table 2 reveals that the mean sun-glint region intensity values vary by up to 6 and 5.7% in scenes one and two, respectively, while the intensity variances in the non-glint regions are essentially zero. In scene one, the DOLP mean values vary by up to 3.2 and 0.14% in the glint and non-glint regions, respectively, while the corresponding variances in scene two are 1.8 and 0.16%, respectively. In general, both the I and DOLP mean values vary more with time in the sun glint regions and follow fluctuation patterns that change depending on the time statistic used for analysis.

2.3 Summary

The simulation results discussed in Subsection 2.1 indicate that the position and number of sun glints within a dynamic sea scene vary continuously over time, although under constant imaging conditions the number of glints does not fluctuate significantly. It was found that the radiation and polarization characteristics of the sun glint regions also fluctuate, with mean I and DOLP values that fall within specific fluctuation ranges that depend upon the time statistic used for assessing the data. In general, the polarization characteristics of sun glint vary continuously, which suggests that sun glint has a time-domain polarization characteristic driven by changes in position, number, and DOLP of sun glint over time. These changes appear not to evolve over time but rather to remain within a specific range.

3. Sea surface clutter suppression method based on time-domain polarization characteristics of sun glint

Because sun glint is primarily horizontally polarized, filtering can be achieved by placing a polarizer in a vertical orientation in front of the detector. In this case, sun glint with higher DOLP produces a better filtering effect within the FOV. The DOLP of sun glint is related to many factors, including the orientation of the sun and the direction of observation. Because the DOLP of sun glint within a FOV is usually below 1, the effect of polarizer filtering is limited and gray saturation will occur even if only a small component of the sun glint is left in the detector.

It was shown in Section 2 that the intensity characteristics of sun glint fluctuate in the time domain and that, over a given period of time, pixels within a fixed scene will shift between states of no glint, some glint, and full glint. Using this feature, Scholl and Gerace assigned the minimum intensities of each pixel in a series of video frame images to a final image, i.e., they defined the desired signal as the non-glint (or minimum glint) signal at each pixel [22]. This method has been experimentally shown to be capable of suppressing sun glint on a water surface, but the approach cannot do so completely and also leaves artifacts and removes some information from the image.

In cases in which a large number of strong sun glints are present in the FOV, neither of the above methods can effectively suppress sun glint. However, we will show below that, by taking the time domain radiation and polarization characteristics of glint into account, it is possible to develop a more effective polarizer filtering method.

3.1 The complete method

The proposed method uses an infrared polarization imaging system that replaces a fixed polarizer with a rotating polarizer that is given specific angular displacements over fixed time intervals. The analyzer angles are evenly distributed over the half-circle (Eq. (5)). At each set analyzer angle, a short-term video image sequence is taken and then processed as follows. First, the minimum value of each pixel in the image sequence is taken to form a preliminary clutter suppression image to be used as the clutter suppression image at the corresponding analyzer angle $f_{θ_{n}} (x, y)$ (Eq. (5)). The clutter suppression images at each analyzer angle are then combined into a sequence image, and the minimum value of each pixel of the combined sequence image is taken to form a final clutter suppression image $F (x, y)$ (Eq. (6)). The data processing process is shown schematically in Fig. 5.

Fig. 5 Schematic of sea clutter suppression method data processing approach.

Download Full Size | PDF

f_{θ_{n}} (x, y) = \min {f_{θ_{n}} (x, y, t)}, θ_{n + 1} - θ_{n} = 180^{\circ} / N,

F (x, y) = \min {f_{θ_{n}} (x, y)}

The steps and procedures outlined above are justified as follows:

a) The use of a rotatable polarizer for imaging at varying analyzer angles helps to suppress clutter. Following Malus’ Law, the intensity of the reflected light from the sea surface reaching the detector is reduced to [23] $I (θ) = I_{p o l} (θ) + \frac{1}{2} I_{n} = I_{p o l} \cos^{2} (\partial - θ) + \frac{1}{2} I_{n},$
where ∂ is the polarization angle, θ is the analyzer angle (the angle of linear polarizer rotation with respect to the horizontal direction), I_pol is the intensity of the sum of the polarization components, and I_n is the natural component.
From Eq. (7), I_n is halved and the intensity of the incident polarized light, I_pol (θ), is reduced by a factor depending on the degree of rotation, θ, resulting in a net reduction of the total energy of the incident light along with an analyzer angle-dependent suppression of background clutter.
In the mid-wave band, sun glint is primarily polarized in the horizontal direction. Correspondingly, we let $\partial$ in Eq. (7) be 0° and plot $I (θ)$ as a continuous (smooth) function of $θ$ (Fig. 6) with a minimum value at $θ = 90 °$ and maxima at $θ = 0$ and $180 °$ . Depending on the analyzer angle, a linearly polarization image that is arbitrarily close to the minimum image $I (90 °)$ can be obtained. For example, if N = 3 linear polarization images with analyzer angles of $0$ , $60$ , and $120 °$ are taken, the images at $60$ and $120 °$ will be relatively close to the minimum image $I (90 °)$ .
According to Eq. (7), the minimum value of $I (θ)$ will be obtained at $θ = \partial \pm 90 °$ . As the value of $\partial$ varies by clutter scene, the proposed method adopts the multi-analyzer angle imaging approach to ensure that the value of $I (θ)$ is as close as possible or equal to the minimum value.
b) Video image sequences are captured at each analyzer angle to suppress the natural component of the clutter using the intensity characteristics of the sun glint. As shown in Section 2, a fixed position within a scene can have varying intensities of glint over time, and it is possible to suppress part of the sun glint signal by taking the minimum value of intensity for each pixel in the video image sequence. However, as will be explained in the next subsection, this step is unnecessary and can be omitted in the actual application.
c) The intensity values of the individual pixels in the combined sequence image are minimized to suppress clutter based on the time-domain polarization characteristics of sun glint. Given $I = I_{p o l} + I_{n}$ and $D O L P = I_{p o l} / I$ , Eq. (7) can be rewritten as $I (θ) = I [D O L P \cos^{2} (\partial - θ) + \frac{1}{2} - \frac{1}{2} D O L P],$
from which the ratio of the intensity of light reflected from the sea surface reaching the detector $I (θ)$ to the total intensity $I$ is
$\frac{I (θ)}{I} = D O L P \cos^{2} (\partial - θ) + \frac{1}{2} - \frac{1}{2} D O L P .$
The simulation analysis results in Section 2 revealed that the sun glint DOLP varies within a specific range over time. Taking the simulation scene two as an example, the range of DOLP is [0.4803, 0.4893]. If the range of analyzer angle during an observation period is [0°, 180°], Eq. (9) can be used to reveal the relationship between the DOLP, $θ$ , and $I (θ) / I$ (Fig. 7).

Fig. 6 Light intensity at detector as a continuous function of analyzer angle (∂ = 0°)

Download Full Size | PDF

Fig. 7 $I (θ) / I$ as a function of $θ$ and DOLP

Download Full Size | PDF

It is seen from the figure that $I (θ) / I$ varies over the range [0.2553, 0.7447]. In other words, this method reduces the light intensity to between 25.53 and 74.47% of the total intensity obtained in simulation scene two. By comparison, using the method in [22] reduces the glint intensity in this scene by only 5.7%. Thus, the proposed method can significantly suppress the intensity of the sun glint received by a detector.

To summarize, the proposed method takes advantage of the suppression effect on sun glint that can be obtained using infrared polarizer filtering. Because the polarization characteristics of sun glint change with time, rotating the polarizer by fixed angular intervals can maximize the suppression effect through the construction of modified video sequence images based on the minimal pixel intensity values at each angle and then the minimal pixel values on the combined image to more effectively remove the effects of glint.

3.2 The simplified method

For the complete clutter suppression method proposed in subsection 3.1, smaller analyzer angle intervals and higher numbers of video sequence image frames at each analyzer angle correspond to a longer total time needed to complete the image acquisition process. However, as the sea surface fluctuates the position of the target will change over time. Because the proposed method processes images acquired at different times, such changes in the target position will tend to change both the relative target position and the target shape in the images. Therefore, an overly long image acquisition process can easily produce artifacts and false information in the final clutter suppression image.

It is therefore necessary to shorten the total acquisition time through simplification of the proposed approach. First, the process of acquiring a video sequence image can be simplified. The purpose of taking the minimum value of each pixel in the video image is to preliminarily suppress the clutter; however, this process will produce artifacts and blurring [22]. If minimum processing is instead applied directly to individual analyzer images, the step of acquiring the video sequence can be omitted by taking only one frame image at each analyzer angle. Further simplification can be achieved by increasing the analyzer angle interval to reduce the number of angles. Using only 0, 60, and 120° as the imaging analyzer angles, the resulting linearized information can be used to obtain the Stokes vector, DOLP, and AOP of the target [24], which can be combined with the clutter suppression image to determine the target polarization characteristics.

Based on the above discussion, a simplified method can be described as follows. Within a certain time interval, the infrared polarization imaging system is used to obtain clutter scenes at analyzer angles of 0, 60, and 120°, which together constitute an image sequence. The minimum value of each pixel in this sequence is then used to form the final clutter suppression image. The next section compares the complete and simplified methods in terms of their respective experimental results and provides further examples of the simplified method.

3.3 The feasibility analysis of the methods

Based on time-domain polarization characteristics of sun glint, we propose the complete method. In order to reduce artifacts and blurring, the simplified method is used. The time-domain polarization characteristics of sun glint make the proposed methods feasible for the following reasons.

a) The change in position of sun glint over time is one of the reasons for the time-sharing acquisition of scene images. In the complete method, there are two minimum operations. The minimum operation on the video takes the advantage of the characteristic of the position change of sun glint over time. The same position may or may not appear sun glint at different times, so the minimum operation may screen out the situation where the position does not appear sun glint. The second minimum operation is also based on this reason. In fact, the position of sun glint is different in linearly polarization images obtained at different times, too. Therefore, in the simplified method, we omit the operation of minimizing the video sequence, and achieve the purpose of suppressing the sun glint by minimizing the sequence image composed of linearly polarized images obtained at different times. This may weaken the clutter suppression effect, but the simplified method has a higher degree of execution, so this simplification is feasible. The experiments in next section can illustrate this.
b) The change in number of sun glint over time does not fluctuate significantly is another reason for the simplified method to omit video acquisition. Video image sequences can be obtained in a short period of time, but because of the time required to rotate the polarizer, the total time to obtain three linearly polarized images with different analyzer angles in the simplified method is often longer than that of a video sequence. The stability of the variation of the number of sun glint makes the performance of sun glint in video sequence images with small time interval and in image sequence with large time interval have little difference. Therefore, it can be predicted that the operation of minimizing the image sequence based on the linearly polarization image with a larger time interval can also achieve the effect of suppressing the sun glint.
c) The change in DOLP of sun glint over time makes the methods to suppress different degrees of sun glint. According to the analysis of subsection 3.1 c), the linearly polarization intensity of sun glint passing through the polarizer can be greatly reduced by changing DOLP of sun glint combined with changing polarization analyzer angles. It can be found from Eq. (9) that the smaller the $I (θ) / I$ , the smaller the intensity of sun glint after passing through the polarizer, and the more suppressed of sun glint. Figure 7 shows the different combinations of the DOLP of sun glint and the analyzer angle have different influence on the $I (θ) / I$ . This indicates that the degree of suppression of sun glint is related to the specific combination mode of the DOLP of sun glint and the analyzer angle. The more the number of analyzer angle, the more likely it is to obtain the best combination mode to suppress clutter. Therefore, when both the DOLP of sun glint and analyzer angle change, the proposed methods can provide a variety of combination modes, which have different degrees of suppression on sun glint and increase the ability of the methods to suppress sun glint.

In fact, when sun glint appears at this moment but does not appear at next moment of a fixed position, minimum operation on a video based on intensity information can suppress the sun glint. When the main component of sun glint is polarized light, the linearly polarization image with the vertical direction to the sun glint AOP can suppress the sun glint. When sun glint always appears in the same position or the sun glint is still strong in the linearly polarization image with the vertical direction to the sun glint AOP, the minimum operation on a video can be replaced by the minimum operation on an image sequence composed of linearly polarization images obtained in different moment with different analyzer angles which utilizes the sun glint suppression effect of the combination of the varying DOLP and the varying analyzer angles. It can be found that, the proposed methods combined the advantage of the minimum operation on a video and the polarizer filtering, and they comprehensively utilize the characteristics of the change of sun glint position, the steady change of the number of sun glint, and the change of the DOLP of sun glint, so the proposed methods can suppress various degrees of sun glint. Therefore, from the theoretical analysis, the proposed methods are feasible.

4. Experiments

The experimental system comprised a rotatable laboratory polarizer mid-wave infrared polarization imaging system using a stepper motor to rotate the polarizer and driven by software written in LabVIEW to control the rotation angle of the rotary table. This system is described in detail in [6]. Two experimental scenes were analyzed: a Tianjin Neihai River scene with a sea surface buoy as the target, and a scene in the Nanchang River in Beijing with a metal cylinder as the target. The target feature is used for registration.

4.1 The sea surface buoy experiment

In the Tianjin Neihai River imaging experiment, the analyzer angle of the polarizer was advanced from 0 to 180° in fixed intervals of 10°. At each fixed analyzer angle, a 50-frame video with a length of 2 s was acquired. The weather during the experiment was fair, with a temperature of 20 °C and a wind speed of 1.2 m/s. The necessity of collecting a video sequence image was assessed by comparing the results of the complete and simplified methods proposed in Section 3. The clutter scene image and the final clutter suppression images produced by the two methods are shown in Fig. 8, in Figs. 8(a), 8(b), and 8(c), respectively.

Fig. 8 Clutter scene image and clutter suppression images from the sea surface buoy experiment. (a) Clutter scene image. (b) Clutter suppression image obtained using complete method. (c) Clutter suppression image obtained using simplified method

Download Full Size | PDF

Two indicators were used to evaluate the clutter suppression effect. The SCR, or ratio of the difference between the mean target and background clutters to the background mean square error, was used to evaluate the differences between target and clutter in the clutter suppression image. Higher values of SCR correspond to a more prominent target and enhanced suppression of background clutter. The mean square error STD, or mean squared difference of intensity values in the background region, was used to evaluate the ability of the clutter suppression method to suppress the background clutter. Smaller values of STD correspond to a better background clutter suppression effect. The preliminary clutter suppression image SCR and STD values produced at each analyzer angle by the complete method are listed in Table 3; the SCR and STD values for the final clutter suppression images produced by the complete and simplified methods are listed in Table 4.

Table 3. SCR and STD of preliminary clutter suppression image at each fixed analyzer angle produced by complete method.

View Table | View all tables in this article

Table 4. SCR and STD of the final clutter suppression images processed by the complete and simplified methods.

View Table | View all tables in this article

It is seen from Fig. 8(b) that the sea clutter is entirely suppressed by the complete method, with both the sun glint and bright band removed. Although the target is retained, some target information is missing. Furthermore, the distant background buildings have become blurred and there are artifacts and false information (black spots in the sky). Thus, the complete method has an obvious sea clutter suppression effect, but because of its long imaging time and the changing position of sea surface buoy there are background artifacts and blurring. A comparison of the preliminary suppression image SCRs and STDs at each analyzer angle (Table 3) produced by the complete method with those of the final clutter suppression image (Table 4) reveals no significant improvement in terms of SCR, which is larger than only seven (36%) of the preliminary suppression image values. This result does indicate that taking the minimum pixel values directly from a video sequence image can, in some cases, improve the target and clutter contrast. In addition, the STD of the final clutter suppression image is smaller than all of the preliminary suppression image values, indicating that the background clutter suppression effect of the complete method is stronger than that achieved by taking the minimum values of the video sequence images alone. Indeed, the preliminary clutter suppression images all have sporadic cluttering, which, although not necessarily reducing the SCR of the preliminary clutter suppression image, affects the visual perception of the image. The final clutter suppression image produced by the complete method does not have any clutter points, rendering it more conducive to highlighting and recognition of the target.

It is seen from Fig. 8(c) that the simplified method suppresses all sea clutter and completely removes sun glint as well as the bright band. The target is essentially preserved, although there is a small amount of artifact and false information in the background. These results highlight the obvious effect of the simplified method in terms of sea clutter suppression. In Table 4, the SCR and STD values produced by the simplified method are both larger than the corresponding complete method results, indicating that, although the simplified method produces a better contrast between target and clutter, it has a relatively inferior ability to suppress background clutter. However, the final image STD produced by the simplified method is smaller than all of the complete method preliminary suppression STDs, indicating that the simplified method does suppress background clutter to a certain extent.

From the above analysis, a combined method that taking the minimum value of each pixel in the image sequence which obtained when polarizer is placed vertically (i.e. combining polarizer filtering and method in ref [22]), complete method and simplified method can be compared. From the comparison of SCR and STD, it can be found that the SCR of the combined method (analyzer angle at 90° in Table 3) is less than that of the complete and simplified methods, and the STD is larger than them. It indicates that the abilities of improving signal-to-clutter ratio and clutter suppression of the combine method are inferior to the complete and simplified methods. The clutter suppression ability of the complete method is better than that of the simplified method, but the improving SCR ability of the simplified method is better than the complete method. Given its overall capabilities at enhancing target clutter contrast, suppressing background clutter, and retaining scene information, the simplified method is the optimal method in this case. Using the simplified method, the video sequence images required by the complete method can be omitted and the number of analyzer angles can be reduced, which further reduces the duration of image acquisition and enhances the observer's understanding of the scene.

4.2 The metal cylinder experiments

Following the analysis method described in Section 4.1, the simplified method was applied in metal cylinder imaging experiments conducted in the Nanchang River, Beijing. Two separate groups of linearly polarized images at three analyzer angles—0, 60, and 120°—were acquired over intervals of 3 s. The resulting linearly polarized and clutter suppression images are shown in Fig. 9, with the corresponding SCRs and STDs listed in Table 5.

Fig. 9 Metal cylinder experiment linearly polarized and clutter suppression images.

Download Full Size | PDF

Table 5. SCRs and STDs of the images in Fig. 9.

View Table | View all tables in this article

In both sets of linearly polarized results, the detector is saturated because the sun glint is too strong and, as a result, the target signal is completely submerged and cannot be found. However, in the clutter suppression images (Fig. 9) the clutter is essentially suppressed, the target information is retained, the contrast between the target and background is significantly improved, and the target is easily detected. Because the clutter is suppressed, the saturated pixels are removed from the images. The dynamic ranges of the images are stretched and the background details are clear. It is seen from Table 4 that the SCRs of the clutter suppression images are much higher than those of any of the linearly polarized images, while the clutter-suppressed STDs are much lower than those of the linearly polarized images. This fully demonstrates that the proposed simplified method can improve the contrast between target and background and is therefore very effective in sea surface clutter suppression, useful for target detection and recognition, and conducive to observer scene comprehension.

5. Discussion

Despite the simplicity and reduced artifact production obtainable under the simplified method, we still recommend using the complete method proposed in this paper when facing a complex clutter environment. As was shown in the analysis in Subsection 4.1, the operation of taking the minimum value of each pixel in a video sequence image can also suppress the clutter to some extent; given a complex scene, suppressing the clutter to find the target is the primary task. Accordingly, the two methods proposed in this paper should be flexibly selected according to the clutter of the scene and the task required.

Because the proposed method collects polarization images of a scene in a time-sharing manner, artifact problems can arise following data processing. If the position of the target changes over time, processing the data without registration leads to the possibility of target artifact; if, on the other hand, the target is registered, the background is prone to artifact. Shortening the total imaging time can reduce the relative displacement of the target between images taken at different times. Similarly, reducing the imaging FOV so that it is filled by the target can reduce deformation of the target over time. If the influence of background artifact can be ignored, target registration can be performed. In cases in which strong clutter saturates the detector pixels, making the target difficult to find, the proposed method can be used for initial clutter suppression even when the target artifact is caused by lack of registration. On the other hand, because the proposed method utilizes the time-domain polarization characteristics of clutter, it is not applicable to scenes in which the polarization characteristics of the clutter do not change with time.

6. Conclusions

Sun glint is one of the main manifestations of sea clutter and seriously affects sea surface image quality. Based on the characteristics of sun glint, we established a polarization radiation model of the dynamic sea surface and used simulation to model how these characteristics change over time. Based on our results, we proposed that sun glint has time-domain polarization characteristics, which were used to develop a method for the polarization imaging of clutter scenes using images or videos taken at different analyzer angles at different times. Clutter suppression is then achieved by combining these images while applying the minimum operation to the individual glint intensities of each image pixel. Experimental results revealed that the proposed method clearly suppresses clutter, which can improve the contrast between target and background and enhance target detection. A simplified variant of the proposed method that shortens the total operation time while increasing applicability was also proposed.

The proposed method takes advantage of the time variation of the polarization characteristics of sun glint and the relative invariance of the polarization characteristics of a target to suppress clutter while highlighting the target. This represents a new approach not taken by previous studies, which neither analyzed nor utilized the time-domain polarization characteristics of sun glint.

Funding

National Natural Science Foundation of China (NSFC) (61575023).

References

1. A. W. Cooper, E. C. Crittenden Jr., E. A. Milne, P. L. Walker, E. Moss, and D. J. Gregoris, “Dual-band infrared polarization measurements of sun glint from the sea surface,” Proc. SPIE 1687, 176–185 (1992). [CrossRef]

2. A.N. de Jong, B. Piet, W. Schwering, P. J. Fritz, and W. H. Gunter, “Optical characteristics of small surface targets, measured in the False Bay, South Africa,” in Proc. SPIE 2009 (2007), 7300: 730003.

3. J. A. Shaw, “Degree of linear polarization in spectral radiances from water-viewing infrared radiometers,” Appl. Opt. 38(15), 3157–3165 (1999). [CrossRef] [PubMed]

4. A. W. Cooper, E. C. Crittenden Jr., E. A. Milne, P. L. Walker, E. Moss, and D. J. Gregoris, “Mid- and far-infrared measurements of sun glint from the sea surface,” Proc. SPIE 1749, 176–185 (1992). [CrossRef]

5. H. Zhao, Z. Ji, Y. Zhang, X. Sun, P. Song, and Y. Li, “Mid-infrared imaging system based on polarizers for detecting marine targets covered in sun glint,” Opt. Express 24(15), 16396–16409 (2016). [CrossRef] [PubMed]

6. J. A. Liang, X. Wang, Y. J. Fang, J. J. Zhou, S. He, and W. Q. Jin, “Water surface-clutter suppression method based on infrared polarization information,” Appl. Opt. 57(16), 4649–4658 (2018). [CrossRef] [PubMed]

7. R. C. Gonzalez and R. E. Woods, Digital Image Processing, 2nd Edition (Prentice Hall, 2003).

8. Y. Xiong and P. Jiaxiong, “An Effective Method for Trajectory Detection of Moving Pixel-sized Target, in Proceed of IEEE Inter Confer on Systems,” Man and Cybernetics, Vancouver, Canada 3, 2570–2575 (1995).

9. M. Diani, A. Baldacci, and G. Corsini, “Joint striping noise removal and background clutter cancellation in IR naval surveillance systems,” IEE P-Vis, Image Sign 148(6), 407–412 (2001). [CrossRef]

10. A. Tartakovsky and R. Blazek, “Effective adaptive spatial-temporal technique for clutter rejection in IRST,” Proc. SPIE 4048, 85–95 (2000). [CrossRef]

11. A. G. Tartakovsky and J. Brown, “Adaptive spatial-temporal filtering methods for clutter removal and target tracking,” IEEE T Aero Elec Sys 44(4), 1522–1537 (2008). [CrossRef]

12. S. Kim and J. Lee, “Small infrared target detection by region-adaptive clutter rejection for sea-based infrared search and track,” Sensors (Basel) 14(7), 13210–13242 (2014). [CrossRef] [PubMed]

13. J. A. Shaw and M. Vollmer, “Blue sun glints on water viewed through a polarizer,” Appl. Opt. 56(19), G36–G41 (2017). [CrossRef] [PubMed]

14. M. Ottaviani, C. Merck, S. Long, J. Koskulics, K. Stamnes, W. Su, and W. Wiscombe, “Time-resolved polarimetry over water waves: relating glints and surface statistics,” Appl. Opt. 47(10), 1638–1648 (2008). [CrossRef] [PubMed]

15. D. K. Lynch, D. S. P. Dearborn, and J. A. Lock, “Glitter and glints on water,” Appl. Opt. 50(28), F39–F49 (2011). [CrossRef] [PubMed]

16. G. Wang, J. Wang, Z. Zhang, and B. Cui, “Performance of eliminating sun glints reflected off wave surface by polarization filtering under influence of waves,” Optik (Stuttg.) 127(5), 3143–3149 (2016). [CrossRef]

17. C. Cox and W. Munk, “Measurement of the roughness of the sea surface from photographs of the sun’s glitter,” J. Opt. Soc. Am. 44(11), 838–850 (1954). [CrossRef]

18. C. Cox and W. H. Munk, “Slopes of the sea surface deduced from photographs of sun glitter,” Scripps Inst Oceanogr Bull 6, 401–487 (1956).

19. M. Ottaviani, C. Merck, S. Long, J. Koskulics, K. Stamnes, W. Su, and W. Wiscombe, “Time-resolved polarimetry over water waves: relating glints and surface statistics,” Appl. Opt. 47(10), 1638–1648 (2008). [CrossRef] [PubMed]

20. S. He, X. Wang, R. Xia, W. Jin, and J. Liang, “Polarimetric infrared imaging simulation of a synthetic sea surface with Mie scattering,” Appl. Opt. 57(7), B150–B159 (2018). [CrossRef] [PubMed]

21. J. Tessendorf, “Simulating Ocean Water,” Presented at SIGGRAPH 2002 course “Simulating Nature,” Realistic and Interactive Techniques,” 21 July, 2002.

22. V. Scholl and A. Gerace, “Removing glint with video processing to enhance underwater target detection,” Presented at 2013 IEEE Western Image Processing Workshop (WNYIPW), New York, 2013. [CrossRef]

23. D. Clarke and J. F. Grainger, Polarized Light and Optical Measurement (Pergamon, 1971)

24. J. S. Tyo, D. L. Goldstein, D. B. Chenault, and J. A. Shaw, “Review of passive imaging polarimetry for remote sensing applications,” Appl. Opt. 45(22), 5453–5469 (2006). [CrossRef] [PubMed]

	Number of effective pixels	Occurrence number of sun glints and ratio to number of effective pixels
	Number of effective pixels	1	ratio	2	ratio	3	ratio	4	ratio
Scene one	24,744	2,223	9%	146	0.6%	10	0.04%	1	0.004%
Scene two	24,713	1,448	5.9%	68	0.28%	4	0.016%	0	0

	Range of intensity mean value		Range of DOLP mean value
	Scene one	Scene two	Scene one	Scene two
Sun glint region	[0.9365,1]	[0.9431,1]	[0.4773,0.4933]	[0.4803,0.4893]
Non-glint region	[0, 4e-07]	[0, 3.7e-07]	[0.0550,0.0557]	[0.0551,0.0559]

	0°	10°	20°	30°	40°	50°	60°	70°	80°
SCR	4.83	6.23	4.03	4.52	4.05	4.22	3.79	3.54	3.19
STD	17.72	12.75	17.52	12.96	12.72	13.95	12.90	12.28	12.64
90°	100°	110°	120°	130°	140°	150°	160°	170°	180°
3.94	5.17	6.24	5.90	5.28	5.64	4.92	4.84	5.18	5.66
12.55	12.51	12.39	11.77	12.00	11.99	13.30	12.64	12.78	12.62

	Complete Method	Simplified Method
SCR	4.3	6.46
STD	10.35	10.93

		0°	60°	120°	Clutter suppression image
Group one	SCR	0.10	0.19	0.22	1.09
Group one	STD	15.21	24.54	21.68	11.74
Group two	SCR	0.34	0.33	0.43	2.98
Group two	STD	26.64	39.92	38.60	7.93

Sea surface clutter suppression method based on time-domain polarization characteristics of sun glint

Abstract

1. Introduction

2. Simulation study on time-domain polarization characteristics of sun glint

2.1 Dynamic sea surface infrared polarization radiation model

2.2 Simulation results and analysis

2.3 Summary

3. Sea surface clutter suppression method based on time-domain polarization characteristics of sun glint

3.1 The complete method

3.2 The simplified method

3.3 The feasibility analysis of the methods

4. Experiments

4.1 The sea surface buoy experiment

4.2 The metal cylinder experiments

5. Discussion

6. Conclusions

Funding

References

Cited By

Figures (9)

Tables (5)

Equations (9)

Optics Express