Skip to content, Skip to search

Changes

SNT: Manual

82 bytes added, 18:45, 22 April 2019
Auto-tracing
===Auto-tracing===
If active, SNT uses the [https://en.wikipedia.org/wiki/A*_search_algorithm A* search algorithm] to automatically construct paths between two manually selected points. To manually place each node nodes in a path, toggle this feature OFF.
[[File:SNT-auto-tracing.png|right|thumb|"Auto-tracing" widget]]
A quick way to improve the quality and efficiency of the pathfinding is to enable "Hessian-based analysis", which filters the image for tube-like structures. Two parameters need to should be chosenspecified:* '''Sigma''' - Controls the spatial scale of the filteringfilter. A lower value of sigma will capture small structural details scale structure but is more sensitive to noise. A larger value of sigma will consider larger scale structure, but may be is less sensitive to local shape characteristics. It is generally advisable advised to pick a value which reflects the average radius of the structures you are tracingto be traced.
* '''Maximum''' - The maximum pixel intensity on the ''Tubeness'' image beyond which the cost function for A* search is minimized.
These parameters may be chosen "Manually" (if you already have a quantitative understanding of the image) or "Visually" (generally more intuitive). If you choose "Visually", the dialog will prompt you to click on a representative portion region of the image which has meaningful structure. Once you click there, a 9-square palette will be is generated showing increasing values of sigma (from top-left to bottom-right) applied to that chunk of the image. The palette is an image stack which can be scrolled through with the mouse wheel.
[[File:Sigma-palette-location.png|left|thumb|400px|Representative region in image]]
<div align="right">
</div>
{{clear}}
Experiment with moving different values of the "Adjusted Max" slider to find a setting which captures the important structures you are interested in, while taking care to avoid over or under-saturation. In the former case, the search might consider pixels outside of the tubular structures, giving a less accurate result. In the latter case, the search might take significantly longer or not finish at all, since each pixel will carry greater cost than it shouldis reasonable. It is usually safe to choose a value somewhere near the median. Once you have chosen this value, click on the square in the palette with the sigma that best fits your application (its border will turn green) then press "Apply". The parameters will be transferred to the interface and a Gaussian convolution will be run over the image, which may take some time depending on its size. Once it's done, path-finding pathfinding will occur on the Hessian image. To display this image during tracing, go to {{bc | View | Show Hessian (Tubeness) Image}} from the SNT dialog.
[[File:original-hessian-side-by-side.png|center|thumb|600px|Original (left) and Hessian (right)]]
758
edits