Retinex - Revision history

XavAL at 21:29, 9 December 2019

2019-12-09T21:29:48Z

← Older revision		Revision as of 21:29, 9 December 2019
Line 1:		Line 1:
			<div class="pagetitle">Retinex</div>

	__NOTOC__		__NOTOC__
	Note: This translation of the "Retinex" French page is a working document.		Note: This translation of the "Retinex" French page is a working document.

DrSlony: /* Generalities */

2019-03-15T10:24:22Z

Generalities

← Older revision		Revision as of 10:24, 15 March 2019
Line 12:		Line 12:
	I relied on two documents :		I relied on two documents :
	*"Automatic Image Haze Removal Based on Luminance Component" (Fan Guo, Zixing Cai, Bin Xie, Jin Tang"[http://www.mcs.csueastbay.edu/~grewe/pubs/DistSensorNetworkBook2011/Atmosphere/HazeREmoval_SingleImage_Luminance.pdf]		*"Automatic Image Haze Removal Based on Luminance Component" (Fan Guo, Zixing Cai, Bin Xie, Jin Tang"[http://www.mcs.csueastbay.edu/~grewe/pubs/DistSensorNetworkBook2011/Atmosphere/HazeREmoval_SingleImage_Luminance.pdf]
	*"Retinex Algorithm on Changing Scales for Haze Removal with Depth Map" (Weixing Wang, Lian Xu)"[~~http~~://~~www~~.~~sersc~~.~~org~~/~~journals/IJHIT/vol7_no4_2014/30.pdf~~]		*"Retinex Algorithm on Changing Scales for Haze Removal with Depth Map" (Weixing Wang, Lian Xu)"[https://scholar.google.com/scholar?cluster=13539545875418694074&hl=en&as_sdt=0,5]
	* and from some programming tricks inspired from "2003 Fabien Pelisson ~~<Fabien.Pelisson@inrialpes.fr>~~"		* and from some programming tricks inspired from "2003 Fabien Pelisson"

	The use of Retinex might be beneficial to images processing:		The use of Retinex might be beneficial to images processing:

DrSlony: Removed British-English spelling

2018-07-18T08:44:25Z

Removed British-English spelling

DrSlony at 01:00, 3 February 2018

2018-02-03T01:00:19Z

Show changes

Lebarhon: /* Adjustments accessible with "Setting" */

2017-08-29T08:03:38Z

Adjustments accessible with "Setting"

← Older revision		Revision as of 08:03, 29 August 2017
Line 229:		Line 229:
	** to adjust Retinex action according to the hue, for example to weaken a sky, to boost leaves …		** to adjust Retinex action according to the hue, for example to weaken a sky, to boost leaves …
	** to reduce the action in the overexposed areas of magenta hue		** to reduce the action in the overexposed areas of magenta hue
	** ~~to notice~~ that when the "highlight" method is selected, this curve also acts on the final image chroma.		** it should be noted that when the "highlight" method is selected, this curve also acts on the final image chroma.
	These 3 adjustments (Gamma Retinex, Histogram Equalizer,Hue equalizer), don't act directly on the "Transmission Map" file, but modify its original value.		These 3 adjustments (Gamma Retinex, Histogram Equalizer,Hue equalizer), don't act directly on the "Transmission Map" file, but modify its original value.
	* Gain and Brightness (Offset): used together, they allow to correctly locate the "haze-free" image. You can find help in the information given in the GUI, and place Min close to 0 and Max close to 32768. Gain and offset act on the "haze-free" image and have no action on the "Transmission Map" file.		* Gain and Brightness (Offset): used together, they allow to correctly locate the "haze-free" image. You can find help in the information given in the GUI, and place Min close to 0 and Max close to 32768. Gain and offset act on the "haze-free" image and have no action on the "Transmission Map" file.

Lebarhon: /* To elaborate a "Transmission Map" file obtained by */

2017-08-29T07:46:42Z

To elaborate a "Transmission Map" file obtained by

← Older revision		Revision as of 07:46, 29 August 2017
Line 49:		Line 49:
	== To elaborate a "Transmission Map" file obtained by ==		== To elaborate a "Transmission Map" file obtained by ==

	* making the difference, for luminance only, between each pixel logarithm of the input image and the neighbouring pixels logarithm of the matching gaussian image. The standard deviation (gaussian function) used here is very high - usually in RawTherapee sigma values from 0.5 to 5 are common – here the values range from 10 to 280 (Single Scale Retinex), it is the user ~~choice~~.		* making the difference, for luminance only, between each pixel logarithm of the input image and the neighbouring pixels logarithm of the matching gaussian image. The standard deviation (gaussian function) used here is very high - usually in RawTherapee sigma values from 0.5 to 5 are common – here the values range from 10 to 280 (Single Scale Retinex), it is up to the user to choose.
	* modifying the input image luminance distribution by		* modifying the input image luminance distribution by
	:* applying a gamma before the "Transmission Map" file creation		:* applying a gamma before the "Transmission Map" file creation

Lebarhon: /* Imposed limitations by RawTherapee */

2017-08-29T07:41:58Z

Imposed limitations by RawTherapee

← Older revision		Revision as of 07:41, 29 August 2017
Line 35:		Line 35:

	The basic algorithm impose a reference to the whole image, and not to a crop or a reduction like in the RawTherapee processing. This limitation imposed by the gaussian function causes several consequences:		The basic algorithm impose a reference to the whole image, and not to a crop or a reduction like in the RawTherapee processing. This limitation imposed by the gaussian function causes several consequences:
	* The processing has been shifted from its dedicated place, that should have been close to "Lab Adjustments" - near the beginning of the RawTherapee processing – that is necessarily (except if somebody has an idea to do ~~different~~) a raw process. As a consequence, non raw files (TIFF, JPG, …) can't be processed with this algorithm? This problem should be solved soon (November 2015 ??).		* The processing has been shifted from its dedicated place, that should have been close to "Lab Adjustments" - near the beginning of the RawTherapee processing – that is necessarily (except if somebody has an idea to do differently) a raw process. As a consequence, non raw files (TIFF, JPG, …) can't be processed with this algorithm? This problem should be solved soon (November 2015 ??).
	* The second consequence of this position is that the characteristics of the raw data situated just after demosaicing are very different that the ones situated downstream: no gamma, no gamut limitation, no white balance, no RGB conversion… so we must expect artefacts and poor luminance and colour rendering.		* The second consequence of this position is that the characteristics of the raw data situated just after demosaicing are very different that the ones situated downstream: no gamma, no gamut limitation, no white balance, no RGB conversion… so we must expect artefacts and poor luminance and colour rendering.
	* The third consequence is the system response time which imposes to re-actualize the whole process after each change in the settings.		* The third consequence is the system response time which imposes to re-actualize the whole process after each change in the settings.

Lebarhon: /* Transmission map adjustment – Iterations with Radius, contrast, scale */

2017-08-28T08:43:38Z

Transmission map adjustment – Iterations with Radius, contrast, scale

← Older revision		Revision as of 08:43, 28 August 2017
Line 348:		Line 348:

	=== Transmission map adjustment – Iterations with Radius, contrast, scale ===		=== Transmission map adjustment – Iterations with Radius, contrast, scale ===

			[[File:TMap.jpg\|300px\|thumb\|right\|Distribution Transmission Map]]
			The next step consist to study in the tool "settings" the indicator that give partial information about Retinex optimization, the "Transmission-map" file, I quoted "Restore haze-free". You can also, iif you want, view the "Transmission-map" file, (set Process to the "Transmission fixed" option).

			We can see the values: Min=-5799 Max=14069., and also other values: TM min =-9.8, Max=7.7 Mean=8.8 Sigma=4.4 Tm=-13.1 TM=20.7

			Let's have a look to the both values Min=-5799 and Max=14069. That means that the image corresponding to these values and which will be subtracted to the original image, doesn't correspond at all to the usual luminance values that, as a reminder, are between 0 and 32768 in RawTherapee using Lab*. The first thing to do is to use the "Transmission-map" curve to get more relevant values. Be careful, the goal isn't to get Min = 0 and Max = 32768, but more reasonable values. For example, it is possible to lower a little bit the left part of the curve and to lift a little bit the central part. We end up for example to the following adjustments: :[[Media:madeira-first.pp3]]

			==== The Haze-free image ====

			[[File:Hazefree.jpg\|300px\|thumb\|right\|Image Haze Free]]
			Just now I don't modify the values that could have an influence on "Transmission-map" like "radius", "contrast".

			Then, in an iterative way between "Gain transmission" + "offset" and "radius", "contrast" and "scale", I will try:
			* to act on the "Transmission-map" curve and "offset" in order to get Min and Max values close to 0 and 32768, but -400 and +39000 may be a good choice too.
			* and then to act on "radius", "contrast", "scale" to obtain a looking fine image, contrasted enough, but without too much effect of depth… it is a matter of taste.
			* work by iterations

			I arbitrary took the following values given in the pp3 [[Media:madeira-second.pp3]]

			=== To apply a Gaussian mask and a gamma ===

			These two tools allow to modify the distribution of the action between high and low lights.
			* choose "Mask Method" = Gaussian mask and increase hihlight = 23 to lighten the sky, you can use the "mask" option to view the changes
			* put "Gamma" in "settings" on "Low"

			Again, all is matter of taste.

			We end up, for example to the following adjustments: [[Media:madeira-third.pp3]]

			== Act on the chroma ==

			Two cases:
			* In the version "Retinex at the beginning of the processing" there is no Retinex action on the chroma. You can use the "Lab adjustments" tool and act either on the "Chromacity" slider or on the two curves a=f(a) and b=f(b), or on the curve C=f© more intuitive.
			* In the version "Retinex in Wavelets", act on the Chroma slider in Wavelet.

			We end up to the following adjustments: [[Media:madeira-four.pp3]]

			== To boost the contrast and to reduce the atmospheric veil even more – Wavelet ==

			* Slightly increase the contrast with "contrast levels"
			* In Residual image" increase contrast and chroma (contrast = 25, chroma = 22), then choose "Compression method" = "Tone mapping" and reduce compression strength = -0.29
			* In "Final Touchup", choose "Contrast balance" select "slider" and adjust "Contrast balance d/v-h" to 19
			* Enable "chroma balance"

			You end up to the following adjustments: [[Media:madeira-five.pp3]]

			== Final adjustment ==

			You can, if necessary :

			* Adjust "Strength" in Retinex
			* Act on "Final contrast" and/or "after contrast curve" in wavelet
			* And also on all the other usual adjustments.

			You end up to the following adjustments: [[Media:madeira-six.pp3]]

Lebarhon at 13:22, 25 August 2017

2017-08-25T13:22:01Z

← Older revision		Revision as of 13:22, 25 August 2017
Line 308:		Line 308:
	\|-		\|-
	\|}		\|}

			= Use in combination with wavelets and Lab Adjustments =

			In case of atmospheric veil, the contrast restoration with Retinex algorithm:
			* is sometimes not good enough or leads to artefacts if "Strength" too high,
			* leads to a chroma and sharpness insufficiency

			Several solutions can be tried, particularly:
			* to increase the chroma using Lab* adjustments (Chroma slider or C/C curve or * = f(a) and b = f(b*) curves;
			* to adjust the global contrast using Lab adjustments (L curve)
			* to increase the chroma using the residual image of wavelet,
			* it is of course possible to use other RawTherapee controls, particularly in the "Exposure" tab.

			But I think that to give priority to the wavelet tool is a good choice, mostly for images with an atmospheric veil.
			We will be, in addition to the adjustments given above, to:
			* increase the residual image contrast,
			* use the "contrast" compression method by sliding down "compression strength" and adjusting "gamma",
			* lightly increase "Edge sharpness" with "edge detection"
			* use "Final touchup", in particular: "Final local contrast", "Final contrast curve" and if necessary ("Tone-Mapping" effect) , "Contrast balance method".

			= A possible example – Dehaze – some recommendations =

			In order to better explain the processing using Retinex, I will take the example of a difficult image, very hazy and tainted with an atmospheric veil [http://rawtherapee.com/shared/test_images/madeira.dng Madeira]

			Be careful, this processing is for pedagogic goal only and has no claim to obtain a perfect image! (by the way, what does it mean, it is subjective). The user has total freedom to proceed to changes, modifications, to get the rendering he wishes.

			== First Retinex adjustments ==

			* At the start [ following not understandable]
			* Enable "Retinex" and that's all

			To get a synthesis of the different adjustments see # Practical synthesis of the various adjustments

			With this kind of image, where the veil is present both in the foreground and in the background and having an uniform histogram, I choose the options:
			* Method: Uniform
			* Lab*: because it is the most efficient of the 3 (Lab, HSL - Log and HSL Lin)

			Then I choose, for "Radius", "Contrast" and "Strength", the adjustments by default, that means respectively 80, 200 and 20.

			=== Transmission map adjustment – Iterations with Radius, contrast, scale ===

Lebarhon at 09:22, 24 August 2017

2017-08-24T09:22:31Z

← Older revision		Revision as of 09:22, 24 August 2017
Line 236:		Line 236:
	* Transmission median filter: reduce the artefacts due to the too strong local variations of "Transmission Map".		* Transmission median filter: reduce the artefacts due to the too strong local variations of "Transmission Map".

	I added a combobox		I added a "combobox" that allow to select the display type. It is both a pedagogic system and also an help to find the "right" settings.
			* Process
			** standard: the default setting
			** mask: display the mask obtained by the Retinex algorithm, resulting from the gaussian processing. We will se here the results of:
			*** the different adjustments placed upstream: method (low, uniform, high and highlight), radius, gamma, histogram equalizer. In the other hand, the other adjustments placed downstream have no impact on the display: contrast, gain, brightness, threshold and the "Transmission map" curve.
			*** the variouos "mask equalizer" adjustments.
			*** you can export these adjustments (TIFF/JPG) to use this mask in external software (Gimp, Photoshop)
			** Unsharp mask: you can use this option in order to subtract the mask to the original image. In this case, the action on "Strength" will alloww to equilibrate the combination between the original image and the mask. Doing so give the possibility to have image with very high radus values.
			** Transmission: display an "image" of the "Transmission Map" file
			*** this "image" doesn't match the reality, that one corresponding to a logarithmic scale which values are generally ranging between -30 to +30.
			*** (fixed): in order to make these values "visible", I applied 2 arbitrary coefficients – I avoided automatic computing to allow comparisons – that shift and boost the basis values.
			*** (auto): I use here the values displayed in the Preview (TM Min Max Mean Sigma) to elaborate a file which values are ranging between 0 and 32700. Of course these data become visible with the maximum range but doesn't match the reality. This configuration may, according to the image, be preferable to "(fixed)".
			*** the displayed image take into account the totality of the Ratinex processings, except a) the "median filter"; b) the "Gain"; c) "Brightness" (offset); d) and of course "Strength".
			*** you can act on the various adjustments mostly the "Mask Equalizer" ones and also the "Transmission Map" curve and thus directly see actions on the haze, the artefacts and the contrast.

			== Practical synthesis of the various adjustments ==

			{\| class="wikitable alternance"
			\|+ Synthesis of the effects of the main adjustments on the contrast
			! Adjustments
			! Main feature
			! Complementary feature - comments
			\|-
			\| Strength
			\| Combine the action of the Retinex image with the original one.
			\| No impact on the Retinex processing
			\|-
			\| Method : Low,Uniforme, High, highlight
			\| Differentiate the action according to the luminance intensity (low = low intensity, highlight = may reduce coloured artefacts)
			\| Higher are the values, more impacted is the foreground (high favours the foreground)
			\|-
			\| Radius
			\| High values for the foreground, low values for the background
			\| High values may create artefacts
			\|-
			\| Contrast
			\| Low values increase the global contrast, favour the foreground, create artefacts
			\| High values reduce artefacts
			\|-
			\| Threshold
			\| Low values increase the global contrast, increase transition artefacts
			\| Low values reduce background artefacts
			\|-
			\| Transmission-map curve
			\| Left part acts on foreground, right part acts on background
			\| Moving the median point acts on contrast and artefacts
			\|-
			\| Gain and Brightness
			\| Locate Retinex according to the source image (see Settings - restored haze-free)
			\| May give illusion to act on contrast and lightness
			\|-
			\|}

			{\| class="wikitable alternance"
			\|+ Other adjustments synthesis
			! Adjustments
			! Main feature
			! Complementary feature - comments
			\|-
			\| Gaussian-mask and Sharp-mask
			\| Reduce the artefacts resulting from too high local contrasts
			\| High values reduce Retinex action
			\|-
			\| Method Lab* - HSL -...
			\| Change the luminance initial repartition, allow to reduce artefacts
			\| Lab* seems more adapted for images with high expositions
			\|-
			\| Gamma and Histogram equalizer
			\| Change the luminance initial repartition to reduce artefacts
			\| Gamma is compensated by its inverse at the end of Retinex processing
			\|-
			\|}

← Older revision		Revision as of 08:44, 18 July 2018
Line 7:		Line 7:
	== Generalities ==		== Generalities ==

	While the eyes are able to see correctly the ~~colours~~ through a poor lighting, a ~~coloured~~ surrounding or a veil of fog, cameras badly manage in theses conditions.		While the eyes are able to see correctly the colors through a poor lighting, a colored surrounding or a veil of fog, cameras badly manage in theses conditions.
	It is by copying the eyes biological mechanisms to adapt itself to these conditions that the MSR algorithm (MultiScale Retinex) has been created. In addition to the digital photography, the Retinex algorithm (Retinex is the contraction for Retina + Cortex) is used in astronomy to show up information laying in the astronomical photographs, in medicine to detect not much visible structures in radiography and tomodensitometry.		It is by copying the eyes biological mechanisms to adapt itself to these conditions that the MSR algorithm (MultiScale Retinex) has been created. In addition to the digital photography, the Retinex algorithm (Retinex is the contraction for Retina + Cortex) is used in astronomy to show up information laying in the astronomical photographs, in medicine to detect not much visible structures in radiography and tomodensitometry.
	Numerous theories and algorithms have been working out for more than 20 years. The first experimentation has been proposed by Rahman in 1996. The approach follows the human visual perception and the Edwin Land's retinal function. In a way, this approach is pretty similar to CIECAM. This function and more particularly its general form is similar to a DOG (Difference Of Gaussian). The idea consists of characterizing the luminous information of a point from its intensity and the intensity of its neighbours. This said, this approach has no scientific basis and is only lying on experience and various empirical constants. Over the years many improvements have been added, but from my own point of view, no one is fully satisfactory.		Numerous theories and algorithms have been working out for more than 20 years. The first experimentation has been proposed by Rahman in 1996. The approach follows the human visual perception and the Edwin Land's retinal function. In a way, this approach is pretty similar to CIECAM. This function and more particularly its general form is similar to a DOG (Difference Of Gaussian). The idea consists of characterizing the luminous information of a point from its intensity and the intensity of its neighbours. This said, this approach has no scientific basis and is only lying on experience and various empirical constants. Over the years many improvements have been added, but from my own point of view, no one is fully satisfactory.
Line 36:		Line 36:
	The basic algorithm impose a reference to the whole image, and not to a crop or a reduction like in the RawTherapee processing. This limitation imposed by the gaussian function causes several consequences:		The basic algorithm impose a reference to the whole image, and not to a crop or a reduction like in the RawTherapee processing. This limitation imposed by the gaussian function causes several consequences:
	* The processing has been shifted from its dedicated place, that should have been close to "Lab Adjustments" - near the beginning of the RawTherapee processing – that is necessarily (except if somebody has an idea to do differently) a raw process. As a consequence, non raw files (TIFF, JPG, …) can't be processed with this algorithm? This problem should be solved soon (November 2015 ??).		* The processing has been shifted from its dedicated place, that should have been close to "Lab Adjustments" - near the beginning of the RawTherapee processing – that is necessarily (except if somebody has an idea to do differently) a raw process. As a consequence, non raw files (TIFF, JPG, …) can't be processed with this algorithm? This problem should be solved soon (November 2015 ??).
	* The second consequence of this position is that the characteristics of the raw data situated just after demosaicing are very different that the ones situated downstream: no gamma, no gamut limitation, no white balance, no RGB conversion… so we must expect artefacts and poor luminance and ~~colour~~ rendering.		* The second consequence of this position is that the characteristics of the raw data situated just after demosaicing are very different that the ones situated downstream: no gamma, no gamut limitation, no white balance, no RGB conversion… so we must expect artefacts and poor luminance and color rendering.
	* The third consequence is the system response time which imposes to re-actualize the whole process after each change in the settings.		* The third consequence is the system response time which imposes to re-actualize the whole process after each change in the settings.
	* The fourth, sometimes, will need a white point modification, in order to avoid ~~colours~~ distortions (for example: a magenta sky).		* The fourth, sometimes, will need a white point modification, in order to avoid colors distortions (for example: a magenta sky).
	* But advantage, this process being situated before "Denoise", the noise reduction will be here fully effective.		* But advantage, this process being situated before "Denoise", the noise reduction will be here fully effective.

	Nevertheless, despite these handicaps, as we will see it further, results are more than satisfactory as well about processing time that in contrast and ~~colours~~ rendering, particularly by joining Retinex and Wavelets actions.		Nevertheless, despite these handicaps, as we will see it further, results are more than satisfactory as well about processing time that in contrast and colors rendering, particularly by joining Retinex and Wavelets actions.

	== Algorithm principle ==		== Algorithm principle ==
Line 213:		Line 213:

	* Strength: Combine "Haze-free" image with the original image. Strength=0 brings no Retinex processing to the image. Strength=100 brings a total processing, "Haze-free" image only is displayed. Values under 50 are recommended.		* Strength: Combine "Haze-free" image with the original image. Strength=0 brings no Retinex processing to the image. Strength=100 brings a total processing, "Haze-free" image only is displayed. Values under 50 are recommended.
	*Chroma (patch): act on the ~~colour~~ component, using the Retinex algorithm, in percentage of the "strength" value relative to the luminance. The slider runs a Retinex process dedicated to the ~~colour~~ component by "simplifying" the process, luminance specific components - gamma, luminance, gaussian mask, ...- aren't executed. Caution, processing times are doubled and the extra memory calls important.		*Chroma (patch): act on the color component, using the Retinex algorithm, in percentage of the "strength" value relative to the luminance. The slider runs a Retinex process dedicated to the color component by "simplifying" the process, luminance specific components - gamma, luminance, gaussian mask, ...- aren't executed. Caution, processing times are doubled and the extra memory calls important.
	*Radius (Neigboring pixels): take into account the neighbouring pixels using the "Difference gaussian" algorithm, higher are the values, more affected is the foreground. Lower are the values, more affected are the distant areas, act directly on the "Transmission Map" file.		*Radius (Neigboring pixels): take into account the neighbouring pixels using the "Difference gaussian" algorithm, higher are the values, more affected is the foreground. Lower are the values, more affected are the distant areas, act directly on the "Transmission Map" file.
	*Highlight threshold: only accessible with the " Highlight" method; set the "Highlight" method threshold, a value of 1 will give about the same effects that the "High" method, high values should allow to reduce the magenta ~~colours~~ in the overexposed areas. Be careful, it will probably be needed to:		*Highlight threshold: only accessible with the " Highlight" method; set the "Highlight" method threshold, a value of 1 will give about the same effects that the "High" method, high values should allow to reduce the magenta colors in the overexposed areas. Be careful, it will probably be needed to:
	** change some adjustments, in particular "Radius" (Neigboring pixels)		** change some adjustments, in particular "Radius" (Neigboring pixels)
	** increase the "Raw white point" value in case of magenta hues in the overexposed areas		** increase the "Raw white point" value in case of magenta hues in the overexposed areas
Line 264:		Line 264:
	\|-		\|-
	\| Method : Low,Uniforme, High, highlight		\| Method : Low,Uniforme, High, highlight
	\| Differentiate the action according to the luminance intensity (low = low intensity, highlight = may reduce ~~coloured~~ artefacts)		\| Differentiate the action according to the luminance intensity (low = low intensity, highlight = may reduce colored artefacts)
	\| Higher are the values, more impacted is the foreground (high favours the foreground)		\| Higher are the values, more impacted is the foreground (high favours the foreground)
	\|-		\|-