1. Introduction

Radar imaging is a category of satellite imagery which creates two-dimensional and three-dimensional images, typically of landscapes. It captures images by imparting its light to illuminate the zone, receiving the reflection and storing the information at the radio wavelength. Radar imaging techniques could be primarily bifurcated into two categories: Real aperture radar (RAR) and Synthetic aperture radar (SAR). RAR captures images by transmitting a pulse of radio wavelength at right angle and further receives the backscattering which gets transformed into an image. While SAR yields high resolution images as it grants distinctive long-term coherency by moving the antenna through a series of positions [1]. It inhibits the movement of radar antenna over a target region to provide finer spatial resolution than conventional beam-scanning radars. The images are configured from the satellite by the rapid interaction and contraction of the emitted radiations from antenna with the target region. This makes SAR produce a finer quality of image over other techniques [3]. Over the last few decades, many researchers have come up with different algorithms and techniques to put forward the most optimal solution for SAR image despeckling [1-6], however there is still a need for a comprehensive study that analyses all of these and giving a high-level view of all the (real speckled SAR image is shown in Figure 1) research that guide the subsequent research in Image processing [15]. This research has come up with a comprehensive and concise study of all the techniques and studies that have been put forward and discussed over the past few decades, which will give the researcher an all-in-one detailed explanation to find their best fit technique [16-18]. Based on research studies the filtering techniques are shown in Figure 2 [19].

Figure 1. Real Speckled SAR image [1].

Figure 2. Filtering techniques [4].

The sections presented in the paper are as follows: The related work is discussed in section 2. Empirical and experimental result analysis of SAR images is studied in section 3. Finally, we procure conclusion and possible future work in section 4.

2. Literature survey

The literature survey of the paper is given in Table 1.

Table 1.

3. Experimental results and discussion

Radar imaging is a category of satellite imagery which creates two-dimensional and three-dimensional images, typically

3.1. Comparative qualitative analysis

In this section, experimental results of the various algorithms studies along with some images are presented to provide evidence for the effectivity of the same. The SAR image used have been shown in Figure 3, while the output of the qualitative analysis is shown in Figure 4. SAR images with artificially added noise of 5% having dimensions of 512 x 512 pixels have been used to test the performance of the algorithms. All experiments have been performed on MATLAB 2018a running on an 8th Gen i5 processor 8GB of Ram and a 64bit system. The impact of the hardware on the performance of the algorithms have been acknowledged. Six of the widely used state of the art speckle noise removal techniques namely, [11][9][7][8][1][14], have been compared against each other. Amongst them,[11] gave the best visual result with a noticeable decrease in the noise and some significant recovery of the details present in the background of the image.

Figure 3. (a) SAR image taken for the evaluation of the techniques (b) Despeckled result [11] (c) Despeckled result [9] (b) Despeckled result [7] (b) Despeckled result [8] (b) Despeckled result [1] (b) Despeckled result [14].

3.2. Comparative quantitative analysis

A set of 12 images were used to measure the accuracy of the techniques, leveraging commonly used image metrics such as SSIM and RMSE. In most of the cases, [11] gave the highest result with [9] at a close second. A set of 12 images were used to measure the accuracy of the techniques, leveraging commonly used image metrics such as RMSE, SSIM.

Table 2. RMSE, SSIM values for the techniques used.

In Table 2 as shown above, the data of the average of all the frameworks across all the techniques have been laid bare for perusal. Form the above data certain inferences can be drawn. The performance of [11][9] and [8] have been very close to each other, with 9% standard deviation. In a similar fashion, [7][14] and [1] have been verisimilitude across the board.

(a)(b)

Figure 5. Plot of average of (a) RMSE and (b) SSIM for the techniques used.

The average values of the scores by the six techniques for the various measures applied over the full dataset are shown in Figure 5. This is done primarily for space reasons since it is beyond the scope of this paper to include 48 graphs to illustrate all techniques across all images. Only in this case [1] performed markedly worse than the others with a score of just 81%. The cause for this is attributed to a change of contrast in all images processed using [1]. RMSE has been taken instead of MSE to normalize the range of variation between the techniquesand between the images themselves. The formula for RMSE is given below by equation 1.

\( RMSE=\sqrt[]{\frac{\sum _{l=1}^{K}||a(l)-\hat{a}(l){||^{2}}}{K}} \) (1)

Here K represents number of data points, a(l) is the l-th measurement, and \( \hat{a}(l) \) is corresponding prediction of \( a(l) \) . In terms of overall quality preservation, both [11] and [9] give similar results, however on magnification of the images it is evident that slight blurring is present at the edges of objects in case of [9]. For practical applications both might give good results, but more evaluation may be needed in special cases. Form the above data certain inferences can be drawn. The performance of [11][9] and [8] have been very close to each other, with 9% standard deviation. In a similar fashion, [7][14] and [1] have been verisimilitude across the board.

4. Conclusion and future scope

In this research, contemporary review on latest SAR image despeckling techniques & approaches is performed including methods and filtering techniques on visual and qualitative perspective namely, Adaptive Speckle Reduction Filter, Conditional Averaging filter, Speckle Reduction Filter, Anisotropic diffusion, Speckle Reduction Filter and Speckle Reduction Filter. These filtering techniques assist in eliminating the noise while preserving the features, details & quality that contribute to significance in radar imagery. Conditional averaging filtering technique and Adaptive Speckle reduction technique despeckle the noise from image by fusion of multiplicative and additive filtering, thus making it significant to restore the vital and necessary information like edges and core that usually settle unfiltered during the initial stages of despeckling. An empirical & theoretical analysis using SSIM, RMSE factors discussing the worthiness of variegated techniques proposed is followed.