Improving Block Matching and 3 Dimensions (BM3D) Filtering for Image Noise Removal Using Discrete Wavelet Transformation (DWT)

In the field of image denoising, Block Matching 3 Dimensions technique (BM3D) stated as a powerful technique that works to remove white Gaussian noise (A WGN). Despite BM3D shows its great success in comparison with the best denoising methods, still its visual quality decreases as the noise density escalates in noisy digital images. Therefore, the need to create a powerful image denoising method is necessary to accomplish the processing of image enhancement applications. The proposed method is introduced by. In this study, discrete wavelet transformation is applied on the LL level of the wavelet decomposition due to its ability to keep the smallest change in the original image. Furthermore, the coefficients of LL will be then divided to small blocks that contain contaminated pixels that suffer from additive white Gaussian noise (A WGN). Comparing with conventional block matching 3D (BM3D) that works well only on the similar patches, our proposed method has the ability to work on similar and dissimilar patches in order to find the inter and intra- correlations between dissimilar pixels. The proposed technique is examined on several digital benchmark images with different noise levels in terms of the quantity of the lost in the original and resulted pixels. In this regard, the performance of the tested images is estimated and then compared proposed with several high quality noise removal techniques according to several assessments such as peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and the figure of merits (FOM). The high amount of these factors reflects the success of the proposed technique. Further, it is interesting to observe the improvement in the running time where it was far less than the investigated noise removal algorithms.