Degrain NodeThe Degrain node is used to help remove grain and other noise from imagery. The default Degrain method is essentially automatic--just connect the Degrain node and you're done. The other available Degrain node operations require more care to use properly, but can result in decreasing the appearance of grain while preserving image detail and sharpness that would be lost otherwise. Choosing a Degrain MethodThe default Degrain method, referred to as Spatial Analysis, performs an area operation not dissimilar to those used for some types of blurs. This averaging of pixel values will reduce the appearance of grain, but it can also eliminate fine image detail. This means that the Spatial Analysis method may make some imagery unacceptably soft. The other two Degrain operations you can choose, both of which use a method called Frequency Analysis, come optimized for film grain or for video noise. Grain and noise are related image phenomena involving high frequency changes in the image content, which tend to be more or less random. Pure noise is completely random, like the snow you get with bad TV reception. Grain is bit more structured, being somewhere between pure noise and structured high-frequency content, such as a long shot of a waving crowd in a stadium. The frequency analysis method examines the image data as a signal. It performs what is known as a coring operation to identify and decrease high frequencies, where both noise and image detail reside. This can also produce an overly soft image, so the Degrain node performs additional operations that compensate for this problem by essentially adding detail and sharpness back into the degrained image. These methods are described in detail below. Degrain Parameter Tab![]() OperationThe Operation parameter enables you to specify the type of analysis used for the degrain operation: spatial or frequency analysis. You can choose between two types of frequency analysis, one optimized for film grain, the other for video noise. Spatial AnalysisThis method uses a median rank operation, which sorts the value of the pixel currently being processed with those of its neighbors and uses the middle value in the ranking. (For more detailed information about this operator, refer to the Rank Node description. The Rank node includes a number of noise reduction presets.) Spatial analysis is the default operation because it is much simpler to use than the other method, frequency analysis. However, the results of spatial analysis are more likely to look soft, or blurry, when compared with the frequency method. Frequency AnalysisFrequency analysis treats the image as a continuous signal of changing intensity (from one pixel to the next and then from one scan line to the next), as it would be represented in the frequency domain. This signal is analyzed to determine which parts of the signal are original, and which are noise or grain. The part determined to be noise, based on the parameters set in the Advanced tab, is removed from the signal. Then this "cleaned up" signal is added back into the original in order to get some or all of the detail back into the image. The frequency analysis approach requires that a number of parameter values be set with care. These parameters are described in the Advanced tab section. Kernel SizeThe Kernel Size menu is used to specify the size of the kernel used by the spatial analysis operation. As the menu options indicate, the smaller 3 x 3 kernel is faster, while the 5 x 5 kernel takes longer to process but produces better results. The frequency analysis method also references the Kernel Size parameter for a blur operation that is performed initially to help identify the high frequency component, which is the difference between the original image and the blurred image. Channel MaskThe Channel Mask parameter enables you to inhibit the modification of any channel by deselecting the Red, Green, Blue, Alpha, or Other channel icons. Control Image ParametersThe Control Action and Control Channel menus enable you to specify how an optional control image input will govern the node operation, as explained in "Using Control Images with Filter Nodes ." Advanced Parameter TabThe parameters on the Advanced tab become available when a frequency analysis option is selected from the Operation parameter on the Degrain tab. However, the channel-specific Scale parameters are disabled when Advanced Video is the selected operation. ![]()
Noise Level and Scale ParametersThese parameters enable you to set a threshold that determines how much of the signal should be subject to filtering. The Noise Level parameter applies to all channels equally, whereas the Scale parameters are used to weight the influence of the noise level value differently for each channel: Noise LevelLarger Noise Level values will remove more of the original signal--they will remove more noise, but also more picture detail. Smaller values will preserve detail, but allow more noise through. Scale (Red, Green, Blue, Other)For film grain analysis, the Scale parameters enable you to adjust threshold values for each channel individually because each film layer responds differently to light. In particular, the blue layer typically has much more noise than the green layer. The default values are a good starting point, as they reflect a ratio typical for many film stocks. (For video, however, the Scale controls are not active, because video signals do not exhibit noise differences between the red, green and blue components.) Gain and Exponent ParametersThese parameters help control how much of the new, degrained signal is added back into the original image: GainHigher values will enhance detail, up to a point; after that, the image will begin to look harsh, or over-sharpened. ExponentThe exponent parameter controls the slope of the nonlinear curve that represents the frequency distribution. In other words, the exponent controls how steep the climb is between low and high:
Typically, more filtering should be done at higher frequencies, which tend to encompass most of both noise and image detail, and less filtering should be done as the frequencies become lower. This relationship is not linear, however, as more of the high frequency material is added than the low frequency information. |