Color noise reduction based on guided filter using luminance information
This application describes method and apparatus for color noise reduction to enhance the quality of color images. An example method first takes a color image as its input and converts it into a Luminance-Chrominance color space, resulting in an YUV image. This YUV representation consists of a Y-channel (Luminance information), which encapsulates the luminance information responsible for the image's brightness, and UV-channels, which represent the chrominance information responsible for the color information. Then a guided filter is created using the Y-channel of the YUV image. The guided filter acts as a reference and assists in refining the subsequent noise reduction steps. With the guided filter in place, the method performs color noise reduction in the UV-channels of the YUV image, where the color noise is primarily present, resulting in smoother and cleaner color values.
1 . A computer-implemented method, comprising:
receiving a color image;
converting the color image into a Luminance-Chrominance color space to obtain an YUV image, wherein the YUV image comprises a Y-channel representing luminance information in the color image, and UV-channels representing chrominance information in the color image;
constructing a guided filter based on the Y-channel of the YUV image;
performing color noise reduction on the UV-channels of the YUV image using the guided filter; and
converting the YUV image with noise-reduced UV-channels into an updated color image with reduced color noise,
wherein:
the color image comprises an RGB (Red, Green, Blue) image, each pixel in the color image comprising a Red value, a Green value, and a Blue value, and
the converting the color image into a Luminance-Chrominance color space to obtain the YUV image comprises:
for a pixel in the color image, computing:
a Y-channel value for the pixel based on the Red value, the Green value, and the Blue value,
a U-channel value based on the Blue value and the Y-channel value, and
a V-channel value based on the Red value and the Y-channel value; and
obtaining the YUV image based on the Y-channel value, the U-channel value, and the V-channel value of the pixel.
2 . The computer-implemented method of claim 1 , wherein the Y-channel value represents brightness information of the pixel, and the U-channel value and the V-channel value represent color difference signals of the pixel.
3 . The computer-implemented method of claim 1 , wherein the constructing the guided filter based on the Y-channel of the YUV image comprises:
constructing a guided image based on the Y-channel of the YUV image; and
wherein the performing color noise reduction on the UV-channels of the YUV image using the guided filter comprises:
defining a filtering window;
for a pixel of the YUV image, computing local Y-channel statistics of the pixel based on pixel values within the filtering window centered at the pixel; and
computing a pair of filtering weights for the pixel based on the local Y-channel statistics of the pixel and UV-channel values of the pixel, wherein the pair of filtering weights are used to perform color noise reduction on the UV-channels of the YUV image.
4 . The computer-implemented method of claim 3 , wherein the pair of filtering weights comprises a first filtering weight and a second filtering weight, and the second filtering weight is a linear function of the first filtering weight.
5 . The computer-implemented method of claim 3 , wherein the constructing the guided image based on the Y-channel of the YUV image comprises:
using the Y-channel of the YUV image as the guided image.
6 . The computer-implemented method of claim 3 , wherein the constructing the guided image based on the Y-channel of the YUV image comprises:
performing brightness noise reduction on the Y-channel of the YUV image to obtain an updated Y-channel representation; and
using the updated Y-channel representation as the guided image.
7 . The computer-implemented method of claim 3 , wherein the constructing the guided image based on the Y-channel of the YUV image comprises:
performing edge detection on the Y-channel of the YUV image to obtain an edge map; and
using the edge map as the guided image.
8 . The computer-implemented method of claim 3 , further comprising:
performing edge detection on the Y-channel of the YUV image to obtain an edge map, and
wherein a first filtering weight of the pair of filtering weights comprises a color noise reduction strength parameter, and the color noise reduction strength parameter is adjustable for individual pixels based on the edge map.
9 . The computer-implemented method of claim 8 , further comprising:
adjusting the noise reduction strength parameter for a pixel based on whether the pixel is within a flat area or an area with an edge.
10 . The computer-implemented method of claim 3 , wherein the performing color noise reduction on the UV-channels of the YUV image comprises:
obtaining filtered pixels by applying the pair of filtering weights to each pixel in the YUV image.
11 . A system comprising one or more processors and one or more non-transitory computer-readable memories coupled to the one or more processors and configured with instructions executable by the one or more processors to cause the system to perform operations comprising:
receiving a color image;
converting the color image into a Luminance-Chrominance color space to obtain an YUV image, wherein the YUV image comprises a Y-channel representing luminance information in the color image, and UV-channels representing chrominance information in the color image;
constructing a guided filter based on the Y-channel of the YUV image;
performing color noise reduction on the UV-channels of the YUV image using the guided filter; and
converting the YUV image with noise-reduced UV-channels into an updated color image with reduced color noise,
wherein the constructing the guided filter based on the Y-channel of the YUV image comprises:
constructing a guided image based on the Y-channel of the YUV image;
wherein the performing color noise reduction on the UV-channels of the YUV image using the guided filter comprises:
defining a filtering window;
for a pixel of the YUV image, computing local Y-channel statistics of the pixel based on pixel values within the filtering window centered at the pixel; and
computing a pair of filtering weights for the pixel based on the local Y-channel statistics of the pixel and UV-channel values of the pixel, wherein the pair of filtering weights are used to perform color noise reduction on the UV-channels of the YUV image.
12 . The system of claim 11 , wherein the pair of filtering weights comprises a first filtering weight and a second filtering weight, and the second filtering weight is a linear function of the first filtering weight.
13 . The system of claim 12 , wherein the constructing the guided image based on the Y-channel of the YUV image comprises:
using the Y-channel of the YUV image as the guided image.
14 . The system of claim 12 , wherein the constructing the guided image based on the Y-channel of the YUV image comprises:
performing brightness noise reduction on the Y-channel of the YUV image to obtain an updated Y-channel representation; and
using the updated Y-channel representation as the guided image.
15 . The system of claim 12 , wherein the constructing the guided image based on the Y-channel of the YUV image comprises:
performing edge detection on the Y-channel of the YUV image to obtain an edge map; and
using the edge map as the guided image.
16 . The system of claim 12 , the operations further comprising:
performing edge detection on the Y-channel of the YUV image to obtain an edge map, and
wherein the first filtering weight comprises a color noise reduction strength parameter, and the color noise reduction strength parameter is adjustable for individual pixels based on the edge map.
17 . A non-transitory computer-readable storage medium configured with instructions executable by one or more processors to cause the one or more processors to perform operations comprising:
receiving a color image;
converting the color image into a Luminance-Chrominance color space to obtain an YUV image, wherein the YUV image comprises a Y-channel representing luminance information in the color image, and UV-channels representing chrominance information in the color image;
constructing a guided filter based on the Y-channel of the YUV image;
performing color noise reduction on the UV-channels of the YUV image using the guided filter; and
converting the YUV image with noise-reduced UV-channels into an updated color image with reduced color noise,
wherein:
the color image comprises an RGB (Red, Green, Blue) image, each pixel in the color image comprising a Red value, a Green value, and a Blue value, and
the converting the color image into a Luminance-Chrominance color space to obtain the YUV image comprises:
for a pixel in the color image, computing:
a Y-channel value for the pixel based on the Red value, the Green value, and the Blue value,
a U-channel value based on the Blue value and the Y-channel value, and
a V-channel value based on the Red value and the Y-channel value; and
obtaining the YUV image based on the Y-channel value, the U-channel value, and the V-channel value of the pixel.
18 . The non-transitory computer-readable storage medium of claim 17 , wherein the constructing the guided filter based on the Y-channel of the YUV image comprises:
constructing a guided image based on the Y-channel of the YUV image;
wherein the performing color noise reduction on the UV-channels of the YUV image using the guided filter comprises:
defining a filtering window;
for a pixel of the YUV image, computing local Y-channel statistics of the pixel based on pixel values within the filtering window centered at the pixel; and
computing a pair of filtering weights for the pixel based on the local Y-channel statistics of the pixel and UV-channel values of the pixel, wherein the pair of filtering weights are used to perform color noise reduction on the UV-channels of the YUV image.