The Study On SAR Image Speckle Suppression And Multi-Sensor Image Fusion

SAR (synthetic aperture radar) images are seriously corrupted by speckle noise, so it is often necessary to enhance the image by speckle suppression before data can be interpreted and analyzed. SAR image carries information quite different from optical image; Image fusion can combine the two types of information together, making them more useful. Based on the above two points, algorithms research in speckle suppression and multi-source image fusion were chosen as the main topic in this paper.1.Speckle noise suppression. The arising mechanism, model and statistical characteristics of speckle noise are described and the popular filters are analyzed. Based on the multiplicative speckle model, a new filtering algorithm based on wavelet transform was developed. First, the multiplicative speckle noise was changed by logarithmic transform into additive noise, which simplified the subsequent speckle suppression. The speckle suppression was carried out in the process of wavelet transform of the logarithmic transformed image. Then the filtered image was achieved by exponential transform. The new filter and some other popular filters have been applied to JERS-1 SAR image. An approach for quality assessment of the filtered images was proposed. After qualitative and quantitative assessment, some conclusions have been drawn:(1)Spatial filters are affected greatly by filtering window size, the size of 5×5 presents best result;(2) Enhanced Frost filter, Enhanced Lee filter and Gamma-MAP filter smooth the speckle noise well with a small sacrifice of losing texture details and narrow edges, which are better than other popular spatial filters;(3) The results show that the proposed filter retains texture and edges well, so more researches are expected.2.Data fusion. After the theories and popular algorithms were reviewed, a new algorithm based on wavelet transform has been proposed and deep research has been carried out to study the impact of the wavelet bases and decomposition levels on the fusion results. After presenting a set of indexes for fusion quality assessment, different fusion algorithms have been applied to three couples of images: JERS-1 SAR image with TM image; ERS-2 SAR image with TM image; SPOT-5 panchromatic image with TM image. With the fusion results been evaluated and compared by the quality assessment indexes presented in this paper, some conclusions have been drawn:(1) For the fusion of SAR image and TM image, compared with the traditional algorithms, the new algorithm based on wavelet transform retains SAR image's texture and structural information well, and its ability to preserve TM image's spectral information is far better than other algorithms;(2)The new algorithm is very flexible due to its various choices of wavelet bases and decomposition levels according to certain applications.(3)Satisfactory results have been achieved by fusing three different types of images with TM image using the new algorithm, which shows that the new algorithm is well adaptive to different types of image data.