Edge Filter

The edge filter effect finds object silhouettes, edges and other areas where contents of the image change rapidly. The edge pixels are set black and other pixels white.

The controls of the effect are:

Threshold Value defines the amount of change between adjacent pixels which is required from edge pixels. The smaller the value, the more edges appear to the image rendered by the effect. A full value 1.0 usually generates an empty white image (when examining the color channel). In geometric filtering, the threshold value is applied to the Distance channel.

Normal Threshold is a second threshold which is used in geometric filtering. It defines the limit angle between two surface normals triggering edge drawing. Threshold value 1 corresponds to the 90 degree angle, 0 to zero angle. A suitably small threshold adds edges also to rounded but tightly bending corners.

Tangent Correction increases distance threshold of geometric filtering at low (tangential) viewing angles. The full correction value 1 is recommended for most scenes.

Threshold Channel selects the data channel, which is used for detecting edges. By default, color changes are examined. You can also use 'Distance' and 'Ray Normal' channels for detecting edges - or any other channel.

Output Channel defines to which channel the black and white result image is written. The default 'Color' channel overwrites the original shaded image. You can write the output to another channel and for example blend it with the shaded output using a VSL effect.

Geometric selects edge detection that is based on geometric continuity. Sharp angles and sudden distance changes generate edge pixels. Threshold channel will not be used.

Geometric outlines

Relative When enabled, distance discontinuity tests are made relative to the size of the scene. Furthermore, the threshold values are automatically adjusted by the output resolution and post image scaling. Distance and normal variations between two adjacent pixels are naturally smaller in a densely sampled high resolution image as in a low resolution preview image. This option automatically compensates the difference and helps to keep the result image similar across output resolution changes. Unfortunately identical detection cannot be guaranteed when resolution changes significantly.

Relative filtering is easier to set up, but the non-relative option provides more accurate control. For example, if a 5 mm thick plate on a table does not show up, reducing the Threshold to a value below 0.005 ensures that the edge will be detected.

[Note] Note
Surface normal based detection needs often some experimentation with the threshold value. An example: a small sphere viewed from a far distance may introduce significant normal direction change between every adjacent pixel. Edge filter which just sees the pixels cannot know if the surface is a 200 face polygonal sphere (=lots of edges) or a perfectly smooth round sphere (=no edges other than the round profile).