|
Abstract:
|
Digital images are degraded by noise either during image acquisition or during image transmission, or both. One of the primary concerns of digital image processing is to reduce this noise for optimal recovery of the original image from a degraded copy. The purpose of this research is to investigate the restoration of images corrupted by signal dependent noise using linear and nonlinear filtering techniques. The type of noise considered in this study is speckle noise which occurs in all types of coherent imaging systems. The linear filtering technique employed here is multiresolution wavelet decomposition of a noisy image and reconstruction of the image by discarding selective detail images at each level of decomposition. We studied this technique by varying the choice of number of wavelet coefficients, and using a synthetic aperture radar (SAR) image and an ultrasound medical image with simulated speckle noise. It was observed that this technique removed the speckle noise of the images appreciably, but caused severe blurring of edges and other sharp details important to them. When the same images were used, the conventional nonlinear median filter removed noise significantly. However, it destroyed some fractal details of the images. A recently developed connectivity preserving morphological filter which is also of the nonlinear type provided the best restoration of images degraded with speckle noise. We characterized the performance of the wavelet, median and morphological filters by using criteria such as visual fidelity, mean square error (MSE), normalized mean square error (NMSE), peak signal-to-noise ratio (PSNR), and intensity profile analysis. |