Attempt to reconstruct color information for pixels that are clipped in one or more RGB channel.

Se questi pixel sono rimasti parzialmente tagliati in uno o due canali, possono risultare in colori non realistici nell’immagine. Per esempio, pixel con i canali verde e blu entrambi tagliati, appariranno rossi senza alcun recupero delle alteluci.

Sono disponibili tre metodi di recupero:

taglio alteluci
blocca tutti i pixel al livello del bianco (es: tagliando i canali colore rimanenti). Questo metodo è il più utile quando il ritaglio (“clipping”) delle alteluci avviene su oggetti desaturati di natura (es: le nuvole).
ricostruisci in LCh
Analizza ciascun pixel con almeno un canale tagliato e tenta la correzione del pixel ritagliato (nello spazio colore LCh) usando i valori degli altri pixel nel blocco interessato nel sensore (3 altri pixel per sensori Bayer o 8 per sensori X-Trans). Le alteluci recuperate rimarranno monocromatiche, ma più luminose e con maggiori dettagli che usando “taglio alteluci”. Questo metodo funziona correttamente con una curva base ad alto contrasto, la quale produce alteluci desaturate. Come “taglio alteluci” questo metodo è una buona opzione per oggetti desaturati di natura.
ricostruisci colore
Usa un algoritmo che trasferisce le informazioni colore dai pixel non ritagliati nelle vicinanze alle alteluci tagliate. Questo metodo funziona molto meglio nelle aree con colori omogenei ed è particolarmente utile sui toni della pelle con alteluci che sfumano dolcemente. Notare che questo metodo può produrre artefatti “a labirinto” sulle alteluci dietro a bordi con alto contrasto, per esempio in caso di strutture correttamente esposte davanti ad uno sfondo sovraesposto.
guided laplacians
Use an algorithm (derived from the diffuse or sharpen module) to replicate details from valid channels into clipped channels and to propagate color gradients from valid surrounding regions into clipped regions. This is a slow and computationally-intensive method designed for maximum smoothness and seamless blending of the reconstructed regions into their neighborhood, and is designed primarily to reconstruct spotlights and specular reflections. This mode is available for Bayer sensors only.

Note: When using the highlight reconstruction included with the filmic rgb module it may be useful to avoid using this module in clip highlights mode (so that filmic rgb has more information to work with).

Controlli modulo

metodo
Il metodo usato per recuperare le alteluci (leggi sopra).
soglia di tosaggio
Regola manualmente la soglia per il tosaggio rispetto alle alteluci magenta. Non dovresti avere bisogno di regolarlo.

Click the icon beside the slider to visualise what areas of the image are considered clipped (the clipping mask). If the clipping mask does not match the RAW over-exposed warning, you may need to correct this value.

noise level (guided laplacians mode only)
Add Poisson noise (natural photon noise such as you would find in sensor readings) to the clipped regions. For high-ISO images, the valid regions of the image will be noisy, but the reconstructed clipped areas will be smooth, which may look odd. Adding some noise in the reconstruction helps to visually blend the result with the rest of the image.
iterations (guided laplacians mode only)
The guided laplacians mode is an iterative process that extrapolates gradients and details from the neighborhood. Each new iteration refines the previous reconstruction but adds more computations that will make the module slower. The default (1 iteration) will typically not be sufficient to completely recover magenta highlights, so you will need to increase this parameter gradually but carefully, to manage the speed/quality trade-off.
inpaint a flat color (guided laplacians mode only)
Inpainting a flat color is an algorithmic trick that may help recover magenta highlights in difficult cases (large blown areas) by smoothing RGB ratios. It can be seen as a “reconstruction booster” that may reduce the number of iterations required to entirely remove magenta in clipped highlights. However, this also makes the reconstruction less accurate and can lead to non-smooth reconstructed edges and unrelated colors being inpainted (e.g. blue sky or green leaves bleeding into white clouds). Use this setting with caution.
diameter of the reconstruction (guided laplacians mode only)
The guided laplacians mode uses a multi-scale algorithm that tries to recover details from each scale independently. The diameter of the reconstruction is the largest scale used by the algorithm. Large scales will increase memory consumption as well as runtimes, and may also cause unrelated colors or details to be inpainted in clipped regions. You are advised to use a diameter roughly twice as large as the largest clipped area to be reconstructed. It is also possible that a given diameter may not suit all clipped areas, in which case you should use several instances at different scales and mask the clipped areas accordingly.

comparison with filmic’s highlight reconstruction

It is important to note that the highlight reconstruction module is quite early in the pixel pipeline – before input color profile and the full chromatic adaptation in color calibration (if you use the modern chromatic adaptation workflow). A common trick to solve clipped highlights is to simply desaturate them to white but, because white is not defined before the full chromatic adaptation and the input color profiling, it is not possible to use this trick here. Technically, there is no color yet at this point in the pipeline, only an arbitrary 3D signal.

The guided laplacians approach has been designed specifically to be immune to white-balance discrepancies and to avoid any concept or method related to color (so there is no explicit desaturation). It only handles gradients (transitions) in the signal and aims at connecting them smoothly, in order to fill the missing parts. This process is quite heavy though, since it falls into the category of supervised machine learning (gradient-based optimization through multi-scale curvature), which is a sub-branch of artificial intelligence.

Filmic’s highlight reconstruction uses a simpler color propagation algorithm coupled with a desaturation option that can favor an achromatic reconstruction. Not only does it know about color (because it comes after the full color profiling and chromatic adaptation) but it also uses a simplified and faster version of the algorithm used by the guided laplacians approach. Namely, this variant will not try as hard to restore details and will favor a smooth blur instead.

The filmic reconstruction is good enough for very large clipped patches and offers the benefit of being able to degrade to white as a last resort. It is also better and faster to inpaint solid color into clipped areas, at the expense of details. Its main drawback is that it is not as selective in the source of the colors being inpainted in clipped parts, so it may inpaint unrelated colors.

All in all, you are advised to use the guided laplacians highlight reconstruction mode to:

  1. smooth the boundaries of clipped areas,
  2. recover spotlights and clipped areas of diameter below approximately 256px (on the full-resolution RAW),
  3. remove chromatic aberrations, which can occur during demosaicing (the next module in the pipeline) at the boundary between clipped and valid regions.

If you find yourself having to increase the diameter of reconstruction past 512px to get a full recovery from magenta, the best approach is usually to cap the diameter to 512px, do the most you can with this setting, and then enable filmic’s highlight reconstruction to finish the work. This will give more bearable run-times with a very similar result.