Image denoising is an essential task in numerous image processing applications. It deals with obtaining a clean image 𝑥 from a noisy image 𝑦 = 𝑥 + 𝑛, where 𝑛 is an additive white Gaussian noise (AWGN). In this thesis, we propose to analyze the behavior of different spatial domain denoising methods, and their effect caused by adaptively chosen weights associated with them. This extensive comparative analysis would assist the researchers to see behavior of such filters in the different circumstances. This will also deal with regional heterogeneity and proposing improvements upon the bilateral filter BF that enhances image recovery. Image denoising using smoothing filters, like BF, is a well-known approach which is based on the weighted average of locally similar pixels. The weights are defined by the spatial and intensity proximity of neighborhood pixels. A limitation of conventional bilateral filter is that its range kernel, which captures intensity similarity of neighborhood pixels, is sensitive to noise. Entropy-based BF (EBF) improves upon the conventional BF by adding more structure to the range kernel and using the entropy image to preselect neighborhood pixels for weighted averaging. The entropy image helps identify heterogeneous regions (quantified by high entropy) and homogeneous regions (quantified by low entropy) in the underlying image which, in turn, allows us to adaptively choose range kernel parameters in EBF. WNNM on the other hand, exploits overall image information by finding similar patches throughout the noisy image and then the matrix of original patches will be a low rank matrix which can be estimated using low rank matrix approximation methods and in the combining all the denoised patched will give us an estimated version of a noise-free image.