flstb_tv TV minimization based on bilinear FLST
flstb_tv [-t scale] [-q quantization] [-p precision] bilinear_tree image
-t scale : Scale (default 5)
-q quantization : gray level quantization (default 1)
-p precision : Curve quantization precision
bilinear_tree : The tree of the interpolated image
image : The output image
void flstb_tv (pScale , pQuantizationLevel , pQuantizationCurve , pTree , pImage )
float *pScale , *pQuantizationLevel ;
int *pQuantizationCurve ;
Shapes pTree ;
Fimage pImage ;
This module reduces the total variation of an image by modifying the local contrast of shapes in a bilinear interpolated image (see flst_bilinear). More precisely, it is a scale-space depending on a scale parameter t. The contrast of a shape relatively to its background is reduced at a speed proportional to the global curvature of its boundary, defined as the ratio C = , L being the length and A the area. The contrast of a leaf in the tree is reduced of tC. If this amount is superior to the original contrast c, the leaf is removed and we try to reduce the contrast of its parent according to the scale t - , provided it has no other child. If it has other children, we apply the same operation to each corresponding branch. If each one is removed in the process, the scale parameter we apply to the shape is t minus the amount of scale used by each branch.
The parameter quantization represents a subsampling of the tree of bilinear level lines such that the contrast between a shape and its parent is at most quantization.
The parameter precision is linked to the discretization of the level lines (see flstb_boundary).
This filter is presented by F. Dibos and G. Koepfler.
flstb_boundary, flstb_dualchain, flst_reconstruct, fsaddles.
Last Modification date : Thu Nov 29 20:23:56 2001