​The Tripartite Pipeline: Alpha Matte Processing

The point of the alpha matte component of the pipeline is to generate an optimised alpha matte that can be forwarded on for color processing.  The following video shows how each node from the above alpha matte part of the pipeline is oriented toward sequentially generating an optimised alpha matte (note red circle around final node and red rectangular frame outline around the actor).  This process is further deconstructed in the following paragraph, which describes the potential use of a shuffle node to dissect the channels of an IMG, and the subsequent following paragraph which details the alpha matte editing process referred to as ‘gamma-slamming’.
Chroma Contrast: The Shuffle Node vs Commercial Packages

Our class footage of a bald guy w face markers, i.e. the foreground of a plate w a green screen background is presenting an IMG where the contrast will (of course, naturally) vary across the 3 different colour channels R, G, and B.   As below, the shuffle node can be used to separate the IMG into the 3 channels as based on chroma contrast  (therez 2 basic types of keying processes, the luma key and the chroma-key; chromakey makes a key based on colour differences).  This process of separating an IMG based on chroma contrast has also been operationalised by means of various commercial packages, such as Keylight or Primatte.
Optimizing the Alpha Matte Contrast: Grade Node and Gamma Slamming
After the IMG was dissected into components as based on color, the resulting channels can be further optimised for contrast by means of various nodes.  The Grade node is used for various refinement strategies; most notable is perhaps the process of increasing contrast by means of optimizing a gamma value (so-called “gamma-slamming”).
​The Alpha Matte Pipeline: Different Foreground, Different PipelineThe preceding Alpha Matte pipeline (for the bald guy w facial trackers) was appropriate for the task required in that instance, but each different foreground will usually involve some degree of variation in the Alpha Matte pipeline strategy.  For example, the following foreground image involves a character who has hair, which thus requires a more elaborate matte to address the edges.  As below, a static image of the overall pipeline is followed by a detailed GIF of the alpha matte nodes relevant to generating an optimized alpha matte.

Picture



​Choking the Matte

Equivalent to the use of gamma slamming in editing an alpha matte, another common method by which the matte is optimised includes choking the matte, i.e. decreasing the size of the white foreground component of the matte as necessary, via the grade node.  Of note, excessive choking of the matte (reducing the gamma values) will introduce pixellated or ‘crunchy’ edges.

​Pipeline, Applied Skills
As an exercise in applied skills, i.e. practicing with node function, the preceding warrior alpha matte was revisited to interrogate the effect of removing a couple nodes.  As apparent from the following screen recording, perhaps there is not much effect from Hair Roto-s and the Edge Alpha Blur Node.  In contrast, upon dissecting the Core Matte, there is an obvious effect from disconnecting the grade node, but disconnecting only the edge blur node (within the core matte) imparted no significant effect.



​Pipeline Organization Revisited

As above, the changes applied to the Warrior pipeline (the original, as provided in class) targeted the issue of whether a couple specific nodes were truly significant in achieving the desired composite matte, or indeed, whether these changes would have any affect on the final composite.  In contrast to the significance of one or two nodes on the visual result of the final composite, of the 2 alpha matte pipelines that were provided in class, it is notable that the bald guy w face markers pipeline could be structured quite differently.  For example, the output from the alpha matte projects a copy node and and  *** node that both reside in the Color Correction pipeline.  This, in itself is not noteworthy, of course the alpha matte must eventually project into the colour pipeline.   It is however notable that the original pipeline (as provided in class) finds the copy output from the alpha pipeline receiving ‘alpha treatment’ in the color pipe.  This arrangement is arguably poor organisation, e.g. the alpha treatment nodes are not implementing a colour based agenda, rather they are refining the alpha matte, and thus presumably belong in the alpha pipe, as per the IMG below.



​Bibliography

Singaraju D, and Vidal R,  (2011).  Estimation of alpha mattes for multiple image layers.  in IEEE Transactions on Pattern Analysis Machine Intelligence. Available at doi 10.1109/TPAMI.2010.206