Filters Overview

All 18 entries sit under Basic Operations Module > Filters. Each takes an Image and returns a Result; the table lists what each computes and the settings that shape it.

Source: StrataQuest 8.1 Engines Reference, ch. 6 (pp. 18–21) and ch. 6.3 (pp. 101–154); User Manual §11.1.2 (p. 159).

Center a small window, the kernel, on one pixel, compute a new value from the pixels under it, then move one pixel over and repeat. In image processing generally, that is neighborhood processing, and most of the Filters list works this way. In StrataQuest the window is usually set by Kernel radius (px): the reference examples pair radius 2 with a 5 × 5 window, radius 3 with 7 × 7 and radius 60 with 121 × 121, so the window is 2 × radius + 1 pixels across.

When the new value is a fixed weighted sum, the filter is a convolutionLoading..., and it is linear: filtering the sum of two images gives the sum of the filtered images. The reference describes Average, Gauss, Laplace, the Sobel filters and Custom filter as filtering with a kernel, and Unsharp filter, an image minus its own blur, is linear too. Median, Filter max and Filter min rank the neighborhood instead, while Bilateral and Anisotropic change their weights with the image, so none of those is linear.

At the image border the window hangs over pixels that don't exist. The module's Options tab has Border Input, which borrows real pixels from neighboring fields of viewLoading... so tiles are filtered as one continuous image, and a Border Option for padding: Fill with black, Fill with white, Mirror or Copy last pixels. Bilateral filter and Sobel filter (cross) also carry their own Border Value and Border type settings.

In image processing generally, smoothing filters are low-pass: they quiet fine detail and keep broad structure. Derivative filters such as Laplace and Sobel are high-pass. Subtracting a low-pass image from the original leaves the high-pass part, which is how Unsharp filter, with a wide kernel, removes a slowly varying background while keeping nucleus-sized detail.

Smoothing. The Average filterLoading... and Gauss filterLoading... blur everything, edges included; Gauss counts near neighbors more. The Median filterLoading... reduces noise while conserving certain borders and makes no new gray levels; in general a median also ignores isolated extreme pixels that would drag an average. To keep edges while smoothing, use the Bilateral filterLoading..., and to smooth along lines and close noise-made gaps in them, the Anisotropic filterLoading....

Edges and detail. The Laplace filterLoading... and the Sobel filtersLoading... return graded images that mark intensity change; follow them with a thresholdLoading... to get a mask. The Canny edge detectorLoading... goes further and returns thin, connected edges from two thresholds. For membranes, the Membrane detectorLoading... detects a double change of intensity along the directions you list, and outputs masks.

Background and sharpness. In StrataQuest the Unsharp filterLoading... is not a sharpener. It subtracts a blurred background model to remove high background, the same job the Background RemovalLoading... engine and the Top-hatLoading... operation do by other routes. For detail and contrast, and for slightly out-of-focus images, the reference points to the Sharp filterLoading....

Extremes and custom kernels. Filter max and Filter min have the same effect as Dilate and Erode in MorphologicalLoading.... Find local max seed and Find local min seed mark regional maxima or minima as a mask, points that, as the names say, can serve as seeds. When nothing built in fits, Custom filterLoading... runs a kernel you type.