Noise-reduction method and apparatus

摘要

A noise-reduction method and apparatus are provided. The noise-reduction method includes estimating activity levels of regions in an input image; and applying different weights to a large noise filter kernel and a small noise filter kernel according to the estimated activity levels, wherein the estimating includes calculating a noise level of the input image, generating a binary image by calculating signal-to-noise ratios (SNRs) of pixels of the input image and by comparing the SNRs of the pixels of the input image with a predefined threshold, and reducing binary data boundaries in the binary image.

主权项

1. A noise-reduction method comprising:
estimating activity levels of regions in an input image, wherein each region comprises a plurality of pixels; filtering the plurality of pixels of a region of the regions in the input image by applying different weights with a large noise filter kernel and a small noise filter kernel according to the estimated activity levels of the regions to obtain a first filtered result of the region, wherein when a weighting of one is applied, output values of the large noise filter kernel and the small noise filter kernel are different than input values of the large noise filter kernel and the small noise filter kernel, respectively, and wherein the estimating of the activity levels of the regions comprises calculating a noise level of the input image, generating a binary image by calculating signal-to-noise ratios (SNRs) of the plurality of pixels of the input image and by comparing the SNRs of the plurality of pixels of the input image with a predefined threshold, and reducing binary data boundaries in the binary image; filtering the plurality of pixels of a region using the large noise filter kernel and the small noise filter kernel with the different weights applied to the large noise filter kernel and the small noise filter kernel, to obtain a first filtered result; and performing filtering of the plurality of pixels of the region using an order statistics filter kernel by applying to the order statistics filter kernel an another different weight that is different from the weights respectively applied to the large noise filter kernel and the small noise filter kernel, wherein the another different weight, which is multiplied with the order statistic filter, is different from weights respectively applied to the noise filter kernel and the small noise filter kernel, wherein the another different weight, which is multiplied with the order statistic filter kernel, is determined according to outlier probabilities of the pixels of the input image, and wherein, when the probabilities of the pixels of the input image being an outlier increases, a weight having a value k applied to the order statistics filter kernel increases and a weight having a value (1−k) applied to the large noise filter kernel and small noise filter kernel decreases.