Differences
This shows you the differences between two versions of the page.


— 
devel:simplify_interest_points [2010/03/31 11:03] (current) mario created 
 ====== Simplification of Interest Point detection and usage====== 
 
 
 ====Current Concept==== 
 
 Currently, for each pair of tiles, the overlapping are is computed, and inside an interest point detection takes place. 
 
 Algorithmic description: 
 * While running xuvstitch, for every tested pair of stacks (stackX, stackY), do: 
 * Compute start and end points of overlapping area: range R 
 * Let stackX be the stack with higher average brightness in range R 
 * Compute voxel gradient magnitude in range R of stackX: gradmagX_R 
 * downscale gradmagX_R by the size of the moving window: scaled_gradmagX_R 
 * find the top N values and their positions in scaled_gradmagX_R: interest_points 
 
 
 ====Proposed Concept==== 
 
 Algorithmic description: 
 * Prior to starting xuvstitch, for every stack (stackX), do: 
 * Compute voxel gradient magnitude of stackX: gradmagX 
 * downscale gradmagX by the size of the moving window: scaled_gradmagX 
 
 The proposed concept will require more memory than the current concept. However, repeated computation of the 'average brightness', the 'gradient magnitudes' and the downscaling is omitted. 
 