Multi Wavelet Transform Domain Image Denoising Using Wiener Filtering and Fuzzy Noise Estimation

The key feature of multiwavelet transform is that it can have properties like short support, orthogonality, symmetry, and vanishing moments, simultaneously, making multiwavelet systems successful in many image processing applications including image denoising. This paper is concerned with image denoising by applying wiener filtering to the approximation coefficients of the multiwavelet transform of the noisy image. However, wiener filtering requires that the noise variance is known. It is conventionally estimated using the mean absolute deviation method, which involves the division by a constant value. When this method is used for different types of images having different statistical features, then the noise estimates become not satisfactory in many cases. Instead of that, a Fuzzy Noise Estimator (FNE) is proposed in this work to be used to provide the required noise variance estimates depending on parameters completely derived from the specific image block.
The proposed image denoising system is simulated and its performance is evaluated in terms of PSNR for thirty five test images belonging to different image classes and having different features. The results show that the proposed system is capable of achieving improvements in PSNR greater than that achieved by two tested conventional multiwavelet based algorithms and the same proposed system but with the FNE is replaced by a conventional noise estimator.