Despeckling of a SAR image without losing features of the image is a daring task as it is intrinsically affected by multiplicative noise called speckle. This thesis proposes a novel technique to efficiently despeckle SAR images. Using an SRAD filter, a Bandlet transform based filter and a Guided filter, the speckle noise in SAR images is removed without losing the features in it. Here a SAR image input is given parallel to both SRAD and Bandlet transform based filters. The SRAD filter despeckles the SAR image and the despeckled output image is used as a reference image for the guided filter. In the Bandlet transform based despeckling scheme, the input SAR image is first decomposed using the bandlet transform. Then the coefficients obtained are thresholded using a soft thresholding rule. All coefficients other than the low-frequency ones are so adjusted. The generalized cross-validation (GCV) technique is employed here to find the most favorable threshold for each subband. The bandlet transform is able to extract edges and fine features in the image because it finds the direction where the function gives maximum value and in the same direction it builds extended orthogonal vectors. Simple soft thresholding using an optimum threshold despeckles the input SAR image. The guided filter with the help of a reference image removes the remaining speckle from the bandlet transform output. In terms of numerical and visual quality, the proposed filtering scheme surpasses the available despeckling schemes.
Date of Award | 2019 |
---|
Original language | English |
---|
- synthetic aperture radar
- image processing
Effective SAR image despeckling based on bandlet and SRAD
Rajamohanan, N. R. (Author). 2019
Western Sydney University thesis: Master's thesis