Stroke Generation Supervisor

About

Our Stroke Generation Supervisor its main task will be to decide which strokes are accepted into the final result and which ones do not meet our quality requirements. See Stroke Accepting On top of that, it will hold some basic settings such as the attempts per segment, the target coverage and more. See General Settings.

Stroke Accepting

We use a Kalman Filter to accept the best strokes on the fly. We dynamically update a normal distribution to match the current color errors. We then use this normal distribution to select the best 50% of the strokes.

Let's plot this Kalman filter over time. In the plot below, you can see the mean of the normal distribution in black. Strokes below the mean are accepted and displayed with green dots. Strokes above the mean are colore red.

We conclude from the figure that without this filter strokes with very high color error would make it to the end result. Thus, the kalman filter indeed is useful.

General Settings

Property Name	Description
max_attempts	Attempts to create a stroke, per segment
supercell_target_coverage	If we have more coverage than this, we wont try fitting.
max_stroke_list_length	The maximum length a list can have.
min_stroke_length_pixels	The maximum length a stroke can have, algebraically.
max_error_threshold	The maximum error a stroke can have with the original image.
min_stroke_coverage_score	The minimum stroke score a stroke must have.
gradient_step_size	The step size for iterative stroke generation.