Edge-Based Noise Cleaning of Chrominance Information in Color Images

It is well known that noise cleaning a full-color image, as if it were a set of independent single-channel images, produces suboptimal results. Instead, it is generally recognized that transforming a color image into a luminance-chrominance space permits the use of simple and aggressive chrominance noise-cleaning operations that minimize degradation of luminance information. These chrominance noise-cleaning techniques usually consist of simple blurring operations that treat all but the lowest chrominance modulations as unwanted noise. While this is effective for noise cleaning high-frequency noise and aliasing artifacts, it can also cause color to bleed across sharp, colored edges in the image. This paper describes a noise-cleaning technique that incorporates edge detection into the chrominance blurring operation in order to prevent color bleeding at image edges. The approach is to first create a map of edge activity in the image and, during noise cleaning, to adaptively increase the noise-cleaning support region for each pixel location based on the local edge activity. By protecting edge fidelity in this manner, noise-cleaning techniques that might otherwise be too aggressive may be applied to the chrominance information. In particular, larger support regions can be used in portions of the image with lower spatial activity. Not only does this increase the effectiveness of reducing colored, highfrequency noise in an image, but it also permits the reduction of lower spatial frequency aliasing artifacts. Introduction One type of noise found in digital camera images appears as low-frequency, colored blobs in regions of low spatial frequency, for example, a person’s face. See Fig. 1. These blobs, a type of chroma noise, produce a mottled appearance in an otherwise spatially flat region. These colored blobs are irregularly shaped and are typically 5 to 25, or more, pixels wide in a given direction. There are numerous existing ways for reducing chroma noise in digital images. Among these are numerous patents that describe chroma noise reduction methods using optical blur filters. These devices frequently address only high frequency chroma noise and are generally ineffective against low frequency chroma noise. Figure 1. Example of chroma noise. Another very common approach to chroma noise reduction is to use standard grayscale image noise reduction techniques on each color channel of the image, in effect, treating each color channel as a separate grayscale image. By treating a full-color image as three, unrelated grayscale images, any interactions or correlations between the color channels are ignored. As discussed below, the inherent relationships between the color planes of a digital image can be used to perform more effective chroma noise cleaning, for example, by transforming the image into a different color space that permits an easier separation of image noise from genuine scene content. Some approaches deal specifically with digital image processing methods for reducing or removing chroma noise artifacts. One class of digital camera patents discloses improvements to the color filter array (CFA) interpolation operation to reduce or eliminate high frequency chroma noise artifacts. Another class of patents, teaches using different pixel shapes (that is, rectangles instead of IS&T's 2003 PICS Conference