Analysis of Speckle Noise Reduction in Synthetic Aperture Radar Images

Analysis of Speckle Noise Reduction in Synthetic Aperture Radar Images

Pranali A. Hatwar Elec. & comm. of Eng. Dept.

A.D. Patel Institute of Technology

New V. V. Nagar, Gujarat-388121

Dr. Heena R. Kher

Elec. & comm. of Eng. Dept.

1. Patel Institute of Technology

   New V. V. Nagar, Gujarat-388121

   Abstract Synthetic aperture radar (SAR) is an active imaging system that can achieve high resolutions both in range and azimuth. Speckle noise is one of the most critical disturbances that alter the quality of SAR coherent images. Before using SAR images in automatic target detection and recognition, the first step is to reduce the effect of speckle noise. In this paper, filtering techniques have been used on the input image and statistical parameters are calculated for the output images obtained from all filters for performance measurement and these are implemented in MATLAB.

   Keywords Synthetic aperture radar (SAR), speckle noise, statistical parameter.

   1. INTRODUCTION

      The idea behind SAR is to synthesize the effect of large aperture by moving small aperture radar along the flight path to obtain much finer resolution. While moving along the flight path, the SAR system records reflected waves from the imaged surface at different instants. Coherent processing of this reflected waves from the illuminated area of different range and azimuth results in the formation of 2-D SAR images [1]. However, such recorded SAR raw data contain unwanted artifacts, which result in granular appearances in SAR image. Those granular parts of SAR image are known as speckle, which is multiplicative in nature and degrades the SAR image quality significantly, leading to loss of crucial information. Therefore, speckle filtering is a critical preprocessing step for many SAR image processing tasks, such as segmentation and edge detection.

      Several methods have been proposed for speckle noise reduction in SAR images. Some methods average out the speckle noise by taking multiple "looks" at a target in single radar sweep [3]. The average in this case is the incoherent average of the look [17]. The other methods involve signal processing using adaptive and non-adaptive filters. Non- adaptive filtering, such as Mean and Median filters, is simpler to implement and requires less computational power, while adaptive speckle filtering is better in preserving edges and details in high-texture areas such as forests and urban areas [8].

      In this paper, the adaptive methods concern for eliminating speckle noise in SAR images. In Section II, gives a brief review of SAR despeckling methods that are present in literature. In section III, the Adaptive despeckling filters and algorithm for speckle noise reduction are discussed for SAR images. In Section IV, estimation of statistical

      parameter for performance measurement is explained. In section V, an experimental result for applying these filters on some real SAR images and compares it with existing speckle removal methods. Finally, in Section VI, we conclude the paper.

   2. A SHORT OVERVIEW OF SAR DESPECKLING METHODS

      Many adaptive filters for SAR image denoising have been proposed in the past. The simplest approaches to speckle reduction are based on temporal averaging, median filtering, and Wiener filtering. Ahmed S. Mashaly et al. [8] proposed Adaptive morphological filter method significantly suppressing speckle noise and preserving the potential targets.

      Recently, there has been considerable interest in using the wavelet transform as a powerful tool for recovering SAR images from noisy data [4]. When multiplicative Conta- mination is concerned; multiscale methods involve a pre- processing step consisting of a logarithmic transform to separate the noise from the original image. Then, different wavelet shrinkage approaches are employed. More specifically, methods based on multiscale decompositions consist of three main steps: First, the raw data are decomposed by means of the wavelet transform, then the empirical wavelet coefficients are shrunk through a thresholding mechanism, and finally, the denoised signal is synthesized from the processed wavelet coefficients through the inverse wavelet transform. Alin Achim et al. [12] suggest method Wavelet-based image de-noising nonlinear SAR. Proposed method based on wavelet decomposition using daubechies and MAP processor is used to estimate signal and noise component instead of thresholding technique. Charles- Alban Deledalle et al. [3] proposed Patch based Non local estimation is used to eliminate speckle with increasing spatial resolution and gives better performance for speckle reduction in SAR intensity image. Method focusing on two methods which are measuring patches similarity and estimate parameters of interest from similar patches. Alin Achim et al.

      [11] suggest method MAP filter Based on the Heavy-Tailed Rayleigh Model. Initially shows the sub-band decomposition of log transformed SAR images can be accurately modeled by alpha-stable distributions. Xie et al. developed a similar method by fusing the wavelet Bayesian denoising technique with Markov-random-field-based SAR image regularization.

   3. ADAPTIVE DESPECKLING FILTERS FOR SAR IMAGES

      The de-speckling filtering is to move a kernel over each pixel in the image and apply a mathematical calculation using the pixel values under this kernel, and then replace the central pixel with the calculated value. The kernel is moved along the image one pixel at a time until the entire image has been covered. In the preceding subsections, we will give a brief review on the mathematical principles of these filters as we will use them in the comparative study with filter.

      The implementation of this filter is based on defining a circularly symmetric filter with a set of weighting values M for each pixel:

      (3.3)

      Where, T is the Euclidean distance between the central pixel of the filter window and its neighbors pixels, while Damp Factor is an exponential damping factor and A is given by:

      The smoothed pixel value given by:

      A. Lee and enhanced-Lee filters

      The lee filter is basically used for speckle noise reduction. This filter is based on the assumption that the mean and variance of the pixel of the interest is equal to the local mean

      R (P1* M1 P2 * M2 … Pn * Mn ) (M1 M2 … Mn )

      (3.4)

      and variance of all pixels with in the moving kernel [20]. The resulting gray level value R for the smoothed pixel is:

      (3.1)

      Where, )

      S= standard deviation of intensity within filter window

      =central pixel of filter window

      =mean intensity with filter window

      The weighting function W is a measure of the estimated noise variation coefficient with respect to the image variation . The number of looks parameter ENL is the effective number of looks of the radar image. ENL is used to estimate the noise variance and it controls the amount of smoothing applied to the image by the filter. Moreover, ENL should be close to the actual number of looks, but it may be changed if the image has undergone resembling. The user may experimentally adjust the ENL value to control the effect of the filter. A small ENL value leads to more smoothing while a large ENL preserves more image features.

      Enhanced-Lee filter divided the SAR image into areas of three classes [16]. The first class corresponds to the homogeneous areas in which the speckles may be eliminated simply by applying a low pass filter, or equivalently, averaging in multi-look processing. The second class corresponds to the heterogeneous areas n which the speckles are to be reduced while preserving texture. The third class is the areas that include the isolated point targets that the filter should preserve. The resulting grey level value R for the smoothed pixel is:

      (3.2)

      Where,W=

      Where, … are grey levels of each pixel in filter window,

      and… are weights computed for each pixel as defined above. The use of large values for damp factor allows better preservation of sharp edges, but reduces the smoothing effect. The use of small values for damp factor increases the smoothing effect, but does not preserve sharp edges well.

      The enhanced Frost filter is similar to the enhanced Lee filter [16] in that it considers three different types of image areas separately: homogeneous, heterogeneous, and areas of isolated point targets. The filter output is:

      (3.5)

      Where, is the result of convolving the image with a circularly symmetric filter whose weighting values M for each pixel is: Where,A=

      Rf (P1* M1 P2 * M2 … Pn * Mn )

      (M1 M2 … Mn )

      C. Kuan and enhanced-Kuan filters

      The kuan filter transforms the multiplicative noise model into an additive noise model [18]. This filter is similar to Lee filter but uses a different weighting function. The resulting grey level value R for the smoothed pixel is:

      (3.6)

      Where,

      The enhanced kuan filter is similar to the enhanced Lee filter in that it considers three different types of image areas separately: homogeneous, heterogeneous, and areas of isolated point targets. The filter output is:

      1. Frost and and enhanced-Frost filters

         Frost filter is designed to smooth out noise while retaining edges or shape features in the image through using an exponentially damped convolution kernel [19]. This kernel adapts itself according to the statistics of the local features.

         Where,

         (3.7)

         1. Root mean square error (RMSE)

            Start

            Mean square error (MSE) is given by:

            N

            MSE

            i j 1

            [f(i, j) F(i, j)]2 / N 2

            (4.2)

            Input SAR image

            Where, f is the original image F is the image denoised with some filter and N is the size of the image.

            Preprocessing of SAR image

            RMSE

            MSE

            (4.3)

            Add Speckle Noise in SAR

            De-noise image output

            RMSE [4] is an estimator in many ways to quantify the amount by which a filtered/noisy image differs from noiseless image.

            Apply De-noising procedure on Degraded SAR image

         2. Peak signal to noise ratio (PSNR)

            PSNR is the ratio between possible power of a signal and the power of corrupting noise that affects the fidelity of its preservation [4].

            PSNR 20log10 (255 / RMSE)

            (4.4)

            Figure 1: Algorithm for Speckle Noise Reduction in SAR Images

         3. Gamma MAP filter

      To apply the MAP (Maximum a posteriori) approach to speckle reduction, the a priori knowledge of the probability density function of the scene is required. With the assumption of a gamma distributed scene, the GMAP filter is derived with the form similar to enhanced Lee [16] with different

      filter model ( ) for heterogeneous areas:

      Higher the PSNR gives lower the noise in the image, i.e. higher image quality.

      D. Correlation Coefficient (CoC)

      Correlation gives the linear relationship between two signals [4] with respect to strength and direction, and its value lies between -1 to +1. The correlation is 1 for increasing linear relationship, -1 for decreasing linear relationship, for all the other cases the value lies between -1 to +1.

      (3.8)

      CoC

      (g g)(g g)

      (g g)2 (g g)2

      (4.5)

      Where,

   4. ESTIMATION OF STATISTICAL PARAMETERS The parameters which are used in the filter performance

      evaluation are Signal to Noise Ratio (SNR), Root Mean Square Error (RMSE), Peak Signal to Noise Ratio (PSNR) and Correlation Coefficient (CoC).

      1. Signal to noise ratio (SNR)

         SNR compares the level of desired signal to the level of background noise. The higher the SNR ratio, the lesser obtrusive the background noise is. SNR in decibels is defined

         as [4]:

         Where, g and are original and images de-noised with some filter respectively, and and are the means of the original image and image de-noised with a few filters respectively.

   5. EXPERIMENTAL RESULTS

      In this section, simulation results are obtained by processing several test SAR images using filtering techniques and we compare the results with speckle filtering methods. In order to be able to quantify the improvement achieved by our method, we have first degraded original noiseless images with synthetic speckle in a controlled manner.

      g e

      SNR 10log( 2 / 2 )

      (4.1)

      Where,is the variance of the noise free image and is the variance of error (between the original and denoised image). Brighter regions have a stronger signal due to more light, resulting in higher overall SNR.

      1. (b) (c)

      Table II: Comparison of different de-noising Filters for SAR image

      gangotri.jpg Corrupted by Speckle Noise with Variance = 0.4

      Filter	SNR (dB)	PSNR (dB)	MSE (dB)	RMSE (dB)	CoC	Elapsed time (sec.)
      Lee filter	5.4843	12.4789	3674.4	60.6171	0.6165	16.317789
      Enhanced Lee	11.8286	18.8232	852.6226	29.1997	0.8877	13.548386
      Frost filter	10.2552	17.2498	1224.9	34.9987	0.8175	35.249355
      Enhanced frost	9.7874	16.7820	1364.2	36.9351	0.7946	36.327350
      Kuan filter	5.4868	12.4814	3672.4	60.6000	0.6166	15.350475
      Enhanced kuan	5.4879	12.4825	3671.4	60.5921	0.6166	15.321556
      Gmap filter	5.4490	12.4436	3704.4	60.8637	0.6130	13.895759

      1. (e) (f)

         (g) (h) (i) Figure 2: Results taken at variance= 0.01 and ENL=3025 on

         SAR image gangotri.jpg. (a) Original image [22] (b) Noisy image (c) Lee filter (d) Enhanced lee filter (damp factor=0.8)

      2. Frost filter (damp factor=0.8) (f) Enhanced Frost filter (damp factor=0.8) (g) Kuan filter (h) Enhanced kuan filter (i)

      Gamma MAP filter

      Table I: Comparison of different de-noising Filters for SAR image

      gangotri.jpg Corrupted by Speckle Noise with variance = 0.01

   6. CONCLUSION

The use of filter in Digital Image Processing improves the image to a great extent. The model preserves the appearances of structured regions. In case of Synthetic Aperture Radar (SAR) Images, Texture and land surfaces have been enhanced. The performance of the algorithm has been tested using statistical parameter measures. Many of the methods are failures to remove speckle noise present in the Synthetic Aperture Radar (SAR) images, since the information about the variance of the noise may not be able to identify by the methods. The Performance of the Speckle noise reduction model for Synthetic Aperture Radar is well as compared to other filters. Lee filters smoothest the image data, without removing edges or sharp features in the images. Enhanced lee filter and enhanced frost filter divide the image into areas of three classes which are homogeneous area, heterogeneous area and isolated point. In the homogeneous area in which speckle eliminated by averaging the multilook processing, in the heterogeneous area in which speckle are reduced while preserving the texture and preserving texture in isolated point target. Kuan, Enhanced kuan and Gamma MAP filter reduces speckle while preserving edges in SAR images.

REFERENCES

1. George Joseph, Fundamentals of Remote Sensing, Universities Press, pp. 1-264, 2005

2. María Elena Buemia, Alejandro C. Frery and Heitor S. Ramos, Speckle reduction with adaptive stack filters,Pattern Recognition Letters 36, pp. 281287, 2014

3. Charles-Alban Deledalle, Loic Denis, Giovanni Poggi, Florence Tupin, and Luisa Verdoliva, Exploiting Patch Similarity for SAR Image Processing,IEEE signal Proc. magazine, pp. 69-78, July 2014

4. K.M. Sharavana Raju, Mohammad Shahnawaz Nasir and T. Meera Devi, Filtering Techniques to reduce Speckle Noise and Image Quality Enhancement methods on Satellite Images, IOSR Journal of Computer Eng., pp. 10-15, Nov. – Dec. 2013

5. Ratil Hasnat Ashique, Md. Imrul Kayes, Speckle Noise Reduction from Medical Ultrasound Images-A comparative study, IOSR Journal of Electrical and Electronics Engineering (IOSR-JEEE), pp. 64-68, Jul.

   – Aug. 2013

6. Signatures, Newsletter of the Indian Society of Remote Sensing

   Ahmedabad Chapter ( ISRSAC), Vol. 24, No.2, Apr-Jun 2012

7. Milindkumar V. Sarode, Prashant R. Deshmukh, Reduction of Speckle Noise and Image Enhancement of Images Using Filtering Technique,International Journal of Advancements in Technology, Vol. 2, pp. 30-38, Jan. 2011

8. Ahmed S. Mashaly, Ezz Eldin F. AbdElkawy, Tarek A. Mahmoud Speckle Noise Reduction in SAR Images using Adaptive Morphological Filter, IEEE 10th Int. Con. on Intelligent Systems Design and Appl., pp. 260-265, 2010

9. Amandeep Kaur, Karamjeet Singh, Speckle noise Reduction by using Wavelets,National Conference on Computational Instrumentation, pp. 198-203, Mar. 2010

10. S.Sudha, G.R.Suresh and R.Sukanesh, Speckle Noise Reduction in Ultrasound Images by Wavelet Thresholding based on Weighted Variance,Internatioaal Journal of Computer Theoryand Engineering, Vol. 1, pp. 7-12, Apr. 2009

11. Alin Achim, Ercan E. Kuruo glu and Josiane Zerubia, SAR Image Filtering Based on the Heavy-Tailed Rayleigh Model, IEEE Trans. on Image Processing, Vol. 15, pp. 2686-2693 Sept. 2006

12. Alin Achim, Panagiotis Tsakalides and Anastasios Bezerianos, SAR Image Denoising via Bayesian Wavelet Shrinkage Based on Heavy- Tailed Modeling, IEEE Transactions on Geoscience and Remote Sensing, vol. 41, pp. 1773-1784, Aug.2003

13. Hua Xie, Leland E. Pierce and Fawwaz T. Ulaby, SAR Speckle Reduction Using Wavelet Denoising and Markov Random Field Modeling,IEEE Trans. On Geosci. and Remote Sensing, vol. 40, pp. 2196-2212, Oct. 2002

14. Samuel Foucher, GozeBertinBenie and Jean-Marc Boucher, Multiscale MAP Filtering of SAR Images,IEEE transactions on image processing, Vol.10, pp. 49-60, Jan. 2001

15. Marc Walessa and MihaiDatcu, Model-Based Despeckling and Information Extraction from SAR Images, IEEE transactions on geoscience and remote sensing, Vol. 38, pp. 2258-2269, Sept. 2000

16. A. Lopes, R. Touzi and E. Nezry, Adaptive speckle filters and scene heterogeneity, IEEE Transactions Geoscience Remote Sensing, vol. 28, pp. 992-1000, 1990.

17. J.S. Lee, Speckle suppression and analysis for synthetic aperture radar, Optical Engineering, vol. 25, pp. 639-643, 1986.

18. D.T. Kuan, A.A. Sawchuk, T.C. Strand and P. Chavell, Adaptive noise smoothing filter for images with signal dependent noise, IEEE Transactions Pattern Analysis And Machine Intelligence, vol. 7, no. 2, pp. 165-177, 1985.

19. V.S. Frost, J.A. Stiles, K.S. Shanmugan and J.C. Holtzman, A model for radar images and its application to adaptive digital filtering of multiplicative noise, IEEE Transactions Pattern Analysis And Machine Intelligence , Vol. 4, no. 2, pp. 157-166, 1982.

20. J.S. Lee, Digital image enhancement and noise filtering by use of local statistics,IEEE Transactions Pattern Analysis and Machine Intelligence, vol. 2, no. 2, pp. 165-186, 1980.

21. http://www.radartutorial.eu/20.airborne/ab08.en.html

22. http://www.nrsc.gov.in/Data_Products_Services_Satellite_Risat1_Sam ple_Images.html