For each blob of the connected component one seed has been selected with the highest residual value. Which, now I think better to chose more than one and probably one seed from lowest residual in the same blob.
Observations: using 3d_CC the objects are better distinguished from each other. We can see that in the flower on the left. However, the features on the butterfly wings could not be preserved as small regions.
0 Comments
3D Regions are color coded after being cut.
3D Poisson disk distribution used for seed selection (radius =12) The first two columns are results from v1 : recoloring processed based on region edges popped from the queue.
The second two columns are results from v2 : recoloring processed based on region's neighbours, using shortest pass. (Please ignore the black holes on the third columns. the problem solved in the 4th column. ) Recoloring starts from reading one region from the queue and recolor the current region and it's neighbours (if they haven't been colored). The process uses the shortest path to recolor all regions.
Here we see the effect of one palette using two different color distance methods. Left to right color distance methods:
CMC: based on the L*C*h color model. Takes RGB as input colors. CIE76: based on Lab color space. Takes RGB colors as input. (L2 norm on Lab parameters) CIE94: defined in the L*C*h* color space with differences in lightness, chroma and hue calculated from L*a*b* coordinates. CIE2000: use the same color space as CIE94 with corrections on perceptual uniformity issue. Euclidean: RGB colors. |
Categories |