An Efficient Algorithm for Image Denoising using NLM and DBUTM Estimation

A desktop study is proposed, to remove noise from original images based on DBUTM (Decision Based Un-symmetric Trimmed Median) estimation method using an NLM (Non Local Means) filter to remove Gaussian and Salt & Pepper noise (impulsive noise). Leaving un-corrupted pixels intact, this filter should be applied to Corrupted Pixels. Without damaging the edges of image for removing impulse noise, Median filters are used. This may lead to hazy and slanted image features. Uncorrupted pixel signal details are eliminated and intensities are altered, while applying the median filter unconditionally to the entire image. For discriminating uncorrupted and corrupted pixels, it is essential to have noise-detection process preceding to applying standard linear filtering method which depends on local spatial correlation. Similar neighbourhood pixels occurring anywhere in the image can be exploited by non-local principles and contributes for denoising using NLM. Decision Based Un-symmetric Trimmed Median (DBUTM) filter is used to identify possible noisy pixels in the image. These pixels are replaced through median filters or its variants leaving other pixels unchanged. Even though image window size is big enough, this filter is an optimum one to detect noise at high level of noises. Experimental results are derived based on various methods namely, Root Mean Square Error (RMSE), Signal to Noise Ratio (SNR), Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM).