An empirical evaluation of wavelets based viz-a-viz classical state-of-art to image denoising

This paper presents an empirical evaluation of wavelet based image denoising methods viz-a-viz classical approaches. Three different types of noises such as salt-&-pepper noise, speckle noise, and Gaussian noise with different variance and mean value have been considered. Wavelet de-noising attempts to remove the noise present in the signal while preserving the signal characteristics. It involves three steps: compute the wavelet transform, use a nonlinear thresholding technique followed by an inverse wavelet transform. The images are decomposed up to level three of HAAR and Daubechies 1 wavelets with two thresholding techniques as soft thresholding and hard thresholding. It was observed that median filter gives better performance over wavelet shrinkage for salt and pepper noise. On the other hand, wavelet shrinkage outperforms over spatial filters for additive Gaussian noise.