![]() |
| 1 |
Описание того, как проходила съемка.
This does not reproduce all images accurately. In the following example, it is actually grossly inaccurate (resulting in black pixels where there should be white): 1. create a new 16x16 image 2. fill with black 3. select 1 pixel brush, paint isolated white dots 4. run wavelet decompose (options do not matter) All pixels where the average difference from their 4 neighbours is >127.5 will be mangled in the output[1]. I realize this is a limitation of grain-merge; Would it be possible to detect this problem and flag it to the user when it occurs? [1] for example, isolated pixels, non-corner parts of L-formations, exact 90degree or 45degree 1pixel lines. 'hat transform': Does this have anything to do with the top-hat transform? http://en.wikipedia.org/wiki/Top-hat_transform I played around with Scipy which includes both black and white top-hat transforms, but was unable to replicate the results.
