I commit a new improvment for locallab with 2 majors enhancements:
a) improve shape detection, especially for : A)burrs when color differences (hue, chroma, luma) are very low, or B) where artifacts appears for example with solid areas with hue (or luma, or chroma) variations
b) improve excludind spot
For the first “improve shape detection”, since one year, I tested 3 solutions
detect structure using “Sobel Canny”, this algorithm is very good, but I don’t be abble to finelized the process… where is the good “line”?
If it is obvious when you see the image, it is not so obvious with the code. After many, many trials, I dropped this algorithm (the code is already present)
detect structure using “Contrast threshold” (as used in sharpening and in combined demosaicing)
I implemanted (experimental) this algo - associated with others - in Color Light curves.
In some cases, it brings someting, but a border appears in some cases, which can be embarrassing, but in most cases, this algorith is not discriminating enough
improves the current algorithm that takes into account actually delta luminance, delta hue, and delta chroma, or a combinaison of this 3 parameters.
With a new algorithm, with others combinations, I think I have arrived at a significant improvement.
Certainly nothing is perfect, but I think the update is appreciable.
For the second point “improve excludind spot”:
- I found two bugs in my algorithm… of course they are solved
- I took advantage of point 3) above to also improve the algorithm
What are the changes ?
- in Color and Light
- some curves types have been improved (for A) and B) above), with 3 experimental choices, once the trials are over, it should only be one, or replace “super”
** Combined hue chroma
** Super + contrast threshold
** Combined hue chroma + contrast threshold
Note that “super” for curves takes into account A)
Note also that sliders take into account B) if checkbox for Lightness - Contrast - chroma “super” is activated
You cannot disabled all this settings.
- for others modules : Exposure, vibrance, blur and noise (including symmetric), tone-mapping retinex, contrast by detail levels, A) and B) are improved but whithout “Contrast threshold”
Module as “sharpening” and “denoise” and “inverse” are not concerned.
I have create in options (manual) a variable “Detectshape”, it is by default “true”. If “true” all modules in 2) above are activated; if “false” RT have the same behavior as previously
When the trial period is over, I will remove this option
To simplify and avoid an extra slider, I used “scope” for each module, it has advantages (simplicity) and disadvantages (not always optimal)
It’s up to the user to find the right value, usually lower for A)…
The variable “Detectshape” is also took into account by “exludind spot”