Despeckle and geographical feature extraction in SAR images by wavelet transform

https://doi.org/10.1016/j.isprsjprs.2007.06.001Get rights and content

Abstract

This paper presents a method to despeckle Synthetic Aperture Radar (SAR) image, and then extract geographical features in it. In this research work, speckle is reduced by multiscale analysis in wavelet domain. In terms of geographical feature preservation the result shows that the method is better compared to spatial domain filters, such as Lee, Kuan, Frost, Ehfrost, Median, Gamma filters. The geographical features such as roads, airport runways, rivers and other ribbon-like shape structures are detected by the new wavelet-based method as proposed by Yuan Yan Tang. Experimental results show that the proposed method extracts geographical features of different width as well as different gray levels.

Introduction

Synthetic Aperture Radar is a kind of high resolution imaging system. It generates images which do not depend on time and weather conditions. It has ability to penetrate through some depth of the soil or vegetation. SAR images are used in many fields, such as agriculture, forestry, geology, hydrology etc. (Fetter et al., 1994). They are inevitably accompanied by speckle due to the coherent nature of the imaging system. The presence of speckle reduces the radiometric resolution of the image and the detectability of the image features. It is usually desirable to reduce speckle prior to image applications. Geographical features such as roads, airport runways and other ribbon-like shape structure detection belong to the category of line extraction or edge detection. Due to speckle the extraction of lines and edges is difficult.

Speckle reduction is becoming a commonly used pre-processing step for geographical feature extraction. The principles of speckle reduction are classified to five categories: 1. control of spatial coherence, 2. control of temporal coherence, 3. spatial sampling, 4. spatial averaging, and 5. digital signal processing (DSP) (Iwai and Asakura, 1996). Many filtering algorithms of DSP category have been developed to reduce speckle such as Lee (Lee, 1980), Enfrost, Kuan (Frost et al., 1982), Median, Gamma, Frost (Kuan et al., 1987), Fukunda and Hirosawa (1998) etc. Most commonly used speckle filters have good speckle-smoothing capabilities. However, the resulting images are subject to degradation of spatial resolution, which can result in the loss of image features (Dong et al., 2001). The amount of speckle reduction must be balanced with smoothness and fine features required in particular applications. For broad-scale interpretation or mapping, fine features can be ignored in many cases. Thus, significant speckle reduction and consequent loss of image features may be acceptable or appropriate in those applications. For applications in which fine features and high resolution are required, the feature preserving performance of a speckle filter is desired.

The work is presented in two parts. In the first part, the speckle is reduced and then features are detected by the new wavelet method (Yang et al., 2003). The wavelet filters have several characteristics:

  • 1.

    They preserve high-frequency information.

  • 2.

    The balance between speckle reduction and detail preservation can be adjusted.

  • 3.

    They require no knowledge of the standard deviation of speckle.

The statistics of the speckle was extensively studied in Goodman (1976), with the conclusion that SAR intensity can be modeled as multiplicative noise. It was proved that for a logarithmically transformed SAR image, the speckle is approximately Gaussian additive noise (Arsenault and April, 1976). Most of the wavelet-based speckle removal approaches can be described in the four steps (Donoho, 1995):

  • 1.

    Take a logarithmic transform on the SAR image (preconditioning).

  • 2.

    Apply the orthogonal discrete wavelet transform (DWT) to obtain the wavelet coefficients.

  • 3.

    Choose a threshold corresponding to the noise variance and apply the soft thresholding.

  • 4.

    Apply the inverse orthogonal DWT and the exponential transform to reconstruct the denoised signal.

This paper is organized as follows. We briefly outline the speckle models of SAR images in Section 2. The basic of wavelet transform is given in Section 3. In Section 4, we outline the proposed method. The performance measurement is presented in Section 5. Some experimental results are presented and discussed in Section 6. Finally, conclusion remarks are given in Section 7.

Section snippets

Speckle models

Coherent imaging technique is widely used in many fields. One of the limitations of this technique is poor image quality affected by speckles. Speckle is produced by coherence interferace related to the roughness of the surface. An accurate and reliable model of speckle is desirable for efficient speckle reduction. The general model is presented in Touzi (2002). It is modeled as multiplicative noise under the assumption that it is fully developed. The input signal f to the linear SAR system is

The Wavelet Transform (WT)

The wavelet transform (Cohen and Kovacevic, 1996) performs the decomposition of a signal into family of functions:ψj,k(x)=2j/2ψj,k(2jxk)generated from a prototype function (mother wavelet) ψ(x) which is dilated by j and translated by k. The mother wavelet has to satisfy the condition given in Eq. (7):ψ(x)x=0.

The discrete wavelet function of a signal f(x) can be computed via a following analysis and synthesis formulas:cj,k=f(x)ψj,k(x)xf(x)=j,kcj,kψj,k(x)where cj,k is approximate

Proposed method

The basic idea of speckle reduction by wavelet thresholding is to convert the multiplicative noise to the case of additive noise. WT-based techniques have proved to be effective due to its compressibility of information signal and incompressibility of noise signal.

The wavelet-based speckle filtering is based on multiscale image analysis. Speckle is suppressed by reducing the amplitude of the pixels in the detail images with horizontal, vertical and diagonal orientation. The flow chart of

Performance evaluation

The performance of the despeckle filter was evaluated and compared with several of the most widely used adaptive filters based on the spatial domain, including the Lee, Frost, Enfrost, Kuan, Median and Gamma filters. The simulation parameters are given in Table 2. For the Gamma filter, only the most commonly used algorithm was used. A common way of estimating the speckle level in coherent imaging is to calculate the mean-to-standard-deviation ratio of the pixel intensity, often termed the

Result and discussion

The single band aerial image of Pilani area (512 × 512 (pixels) × 255 (gray levels)) is taken as original image. It is contaminated by normal distributed multiplicative noise of variance range 0.01 (standard deviation 0.1, 10% noise) to 0.05 (standard deviation 0.223, 22.3% noise) with step size of 0.01. The filter parameters (Dong et al., 2001) used in simulations are specified in Table 2. Three different windows are used to find the local statistical parameters for spatial filtering. Table 3

Conclusion

The experimental results show that the proposed filter is better than several commonly used filters, including Lee, Frost, Lee-Sigma and Gamma-Map, in terms of despeckle and detail preservation. The knowledge of speckle standard deviation is not required in wavelet-based filtering as compared to other most commonly used speckle filters. The estimation of speckle standard deviation is not an easy task. For geographical feature detection, a novel wavelet-based ribbon-like structure recognition

Acknowledgement

The author is thankful to NRSA, Hyderabad, for providing single band aerial image of Pilani area.

References (17)

  • H.H. Arsenault et al.

    Properties of speckle integrated with a finite aperture and logarithmically transformed

    Journal of the Optical Society of America

    (1976)
  • A. Cohen et al.

    Wavelets: the mathematical background

    Proceedings of the IEEE

    (1996)
  • Y. Dong et al.

    Toward edge sharpening: a SAR speckle filtering algorithm

    IEEE Transactions on Geoscience and Remote sensing

    (2001)
  • D.L. Donoho

    Denoising by soft-thresholding

    IEEE Transactions on Information Theory

    (1995)
  • F.M. Fetter et al.

    Sea ice type map from Alaska synthetic aperture radar facility imagery: an assessment

    Journal of Geophysical Research

    (1994)
  • V.S. Frost et al.

    A model for radar images and its application to adaptive digital filtering of multiplicative noise

    IEEE Transactions on Pattern Analysis and Machine Intelligence

    (1982)
  • S. Fukunda et al.

    Suppression of speckle in synthetic aperture radar images using wavelets

    International Journal of Remote Sensing

    (1998)
  • J.W. Goodman

    Some fundamental properties of speckle

    Journal of the Optical Society of America

    (1976)
There are more references available in the full text version of this article.

Cited by (40)

  • Monitoring of flood water propagation based on microwave and optical imagery

    2021, Quaternary International
    Citation Excerpt :

    The backscattering coefficients (in dB) was learnt from the orthorectified sigma0 band by equation (10)*log 10 (sigma0) (Li et al., 2011). Speckle noise was detached by using a single product speckle filter method with a window size of 7*7 pixels (Gupta and Gupta, 2007; Schmidt and Skidmore, 2004; Tian et al., 2015). The median value of all pixels at the same location for each period was used to extract the seasonal surface water extent.

  • LNPSS: SAR image despeckling based on local and non-local features using patch shape selection and edges linking

    2020, Measurement: Journal of the International Measurement Confederation
    Citation Excerpt :

    Some approaches based on wavelet transform texture features are utilized to remove SAR image speckle noise. They not only can decline speckle noise but they also can preserve important edge information in a defined degree [24–27]. In the literature [28,29] the use of dictionaries for learning has indicated notable results in image despeckling.

  • A new metric for measuring structure-preserving capability of despeckling of SAR images

    2014, ISPRS Journal of Photogrammetry and Remote Sensing
    Citation Excerpt :

    According to the multiplicative model of speckle noise, the ratio image should ideally contain noise removed by the despeckling process while the presence of structural information indicates corresponding loss in despeckled results. For this reason, equivalent number of looks (Touzi, 2002; Pellizzeri, 2003; Guptaa and Guptab, 2007; Bianchi et al., 2008; Molina et al., 2010; Hou et al., 2012) of ratio images (ENLR) is adopted whose deviation from the nominal number of looks (NNL) of SAR images is considered as a measure of structure preservation. Although perfect despeckling results make the ENLR with the nominal value, the degree of deviation may not be consistent with the actual loss of structural features.

View all citing articles on Scopus
View full text