Study On The Speckle Reduction Algorithm For Synthetic Aperture Radar Image Based On The Similarity Measure

Synthetic aperture radar (Synthetic Aperture Radar, SAR) has been widely used inmarine and terrestrial observations for its unique advantages of all-time, all-weather,multi-polarization, multi-band and high-resolution imaging. Speckle is an inherentphenomenon in the SAR imaging process. The presence of speckle reduces the readabilityof the image and complicates the image interpretation and analysis. Thus the specklereduction has become an important part of the SAR image processing. In recent years, alarge number of speckle reduction algorithms have been proposed, such as Lee, Frost,Kuan filtering methods which are based on the calculation of statistical properties;anisotropic diffusion and wavelet filtering methods which are based on gradient featuresand other filter methods based on model characteristics. A common problem in the study ofspeckle reduction is that it’s difficult to maintain the edge details of images effectivelywhile the speckles are eliminated. In this paper, speckle reduction for SAR images isstudied. An improved speckle reduction algorithm is proposed with the introduction ofstructural similarity measure. The main work of this paper is as follows:1. An improved Lee filtering algorithm based on structural similarity measure isproposed. The inter-pixel similarity function is adopted in this algorithm. The filteringprocess is guided by the calculation of inter-pixel similarity function which is used todiscriminate structural attributes of the pixels.For the problem of insufficient noiseimmunity at the pixel level, image blocks surrounded the central pixel are defined todistinguish the structural attributes of the pixels firstly. In a greater range of neighborhood,pixels which are the most similar to the center pixel in structural properties are selected.Secondly, pixel similarity function is calculated. The new two-stage Lee filteringalgorithm based on structural similarity measure is proposed. Through this algorithm, theimpact of noise on the image is weakened and the denoising performance is improved.2. The improved Kuan and Frost filtering algorithms based on two-stage structuralsimilarity measure are proposed. The algorithm described above is based on the calculationof statistical properties of the image pixels to measure the similarity between pixels. In thisdissertation, the two-stage structural similarity measure is extended to apply to the Kuanand Frost filtering algorithms. It is adapted to these two filtering algorithms which arealso based on the calculation of statistical properties of the image pixels.The experimental results show that compared with previous Lee filtering algorithms, the improved two-stage structural similarity measure Lee filtering algorithm hassignificant advantage in the edge-preserving while the speckle noise is eliminated. Andapplying this method to the Frost and Kuan filtering algorithms can also be effectivein speckle reduction with better performance on maintaining image details at thesame time.