Role of two-dimensional Gaussian filter in image noise removal

Size of filter plays important role in image noise removal. The present study investigates the effect of filter size on image noise removal. A digital phantom with image matrix size 128 × 128 was created. It consists of eight objects (three squares, one circle, three solitary rectangular bars, and one triangle made up of six rectangular bars, mimicking hot spots) of different sizes. Gaussian noise with variance ranging from 0.003 to 0.15 incremented at the rate 0.003 was added to the phantom image, thus creating 50 noisy images. Gaussian filters of 3, 5, 7, 9, and 11 pixel size were applied to noisy images. The output image quality was assessed both subjectively and objectively using peak signal-to-noise ratio (PSNR). Key At minimum noise level, for filter sizes 3 and 11 pixels, the PSNR value of output image was 20 dB and 15 dB, respectively, while at maximum noise level, it was 15 and 14 dB. Visually, at maximum noise level, better image quality, i.e., sharp image with very minimum edge distortion was noticed with 3-pixel size filter; however, amount of noise removal was least in comparison to 5, 7, 9 and 11 pixel size filter. The maximum noise removal, maximum edge distortion, and elimination of objects of size less than the filter size was found with 11-pixel size filter. Research Better image quality was noticed with 3-pixel size Gaussian filter. (author)