Demosaicing - Revision history

XavAL at 21:14, 9 December 2019

2019-12-09T21:14:42Z

← Older revision		Revision as of 21:14, 9 December 2019
Line 1:		Line 1:
			<div class="pagetitle">Demosaicing</div>

	{{1:1}}		{{1:1}}

DrSlony: /* Introduction */

2019-05-07T09:07:04Z

Introduction

← Older revision		Revision as of 09:07, 7 May 2019
Line 16:		Line 16:
	RawTherapee offers several demosaicing algorithms, each with its own characteristics. The differences between them can be subtle - one might need to zoom in to 100% or more to discern them. However, as the demosaiced image constitutes the foundation upon which all other tools work, the choice of demosaicing algorithm can have a visually significant effect on the end result, particularly when viewed up close. The most visible effects of the choice of demosaicing algorithm include the rendering quality of fine detail and the visibility of artifacts in the form of maze-like patterns.		RawTherapee offers several demosaicing algorithms, each with its own characteristics. The differences between them can be subtle - one might need to zoom in to 100% or more to discern them. However, as the demosaiced image constitutes the foundation upon which all other tools work, the choice of demosaicing algorithm can have a visually significant effect on the end result, particularly when viewed up close. The most visible effects of the choice of demosaicing algorithm include the rendering quality of fine detail and the visibility of artifacts in the form of maze-like patterns.

	Concerning Bayer cameras, AMaZE is generally the best method for lower-ISO images, ~~and~~ LMMSE or IGV are better at higher-ISO ones. Concerning X-Trans cameras, 3-pass (Markesteijn) is generally the best method.		Concerning Bayer cameras, AMaZE is generally the best method for lower-ISO images, while LMMSE or IGV are better for higher-ISO ones. Concerning X-Trans cameras, 3-pass (Markesteijn) is generally the best method.

	On a side note, the [https://en.wikipedia.org/wiki/Foveon_X3_sensor Foveon X3 sensor] does not use a color filter array and so images coming from camera with such a sensor do not need to be demosaiced. They are, however, unsupported by RawTherapee.		On a side note, the [https://en.wikipedia.org/wiki/Foveon_X3_sensor Foveon X3 sensor] does not use a color filter array and so images coming from camera with such a sensor do not need to be demosaiced. They are, however, unsupported by RawTherapee.

DrSlony: /* Dual Demosaic */ typo fixed

2019-05-07T09:06:14Z

Dual Demosaic: typo fixed

← Older revision		Revision as of 09:06, 7 May 2019
Line 114:		Line 114:
	==== Dual Demosaic ====		==== Dual Demosaic ====

	The dual-demosaic methods, such as AMaZE+VNG4, allow you to demosaic areas of high contrast (i.e. detail) using one method and areas of low contrast (i.e. no detail, plain areas such as sky) using the other algorithm. As some algorithms are better ~~are~~ rendering fine detail while others are better at rendering smooth, plain areas, these dual-demosaic options allow you to combine the best of both worlds, this providing a better starting point for the other tools and mitigating the need for sharpening or noise reduction. The downside is that the image needs to be demosaiced twice, thus taking longer than using a single demosaicing method.		The dual-demosaic methods, such as AMaZE+VNG4, allow you to demosaic areas of high contrast (i.e. detail) using one method and areas of low contrast (i.e. no detail, plain areas such as sky) using the other algorithm. As some algorithms are better at rendering fine detail while others are better at rendering smooth, plain areas, these dual-demosaic options allow you to combine the best of both worlds, this providing a better starting point for the other tools and mitigating the need for sharpening or noise reduction. The downside is that the image needs to be demosaiced twice, thus taking longer than using a single demosaicing method.

	The threshold for adjusting the level of detail at which one demosaicing algorithms should be used over the other is controlled using the "contrast threshold" slider. The slider includes a checkbox which computes an optimal level automatically.		The threshold for adjusting the level of detail at which one demosaicing algorithms should be used over the other is controlled using the "contrast threshold" slider. The slider includes a checkbox which computes an optimal level automatically.

DrSlony: /* Introduction */

2019-05-06T14:34:54Z

Introduction

← Older revision		Revision as of 14:34, 6 May 2019
Line 10:		Line 10:
	# There is no concept of color yet, as each photosite registers only a single electric charge induced by the photons which pass through the filter and strike it,		# There is no concept of color yet, as each photosite registers only a single electric charge induced by the photons which pass through the filter and strike it,
	# There are twice as many green photosites as either red or blue,		# There are twice as many green photosites as either red or blue,
	# and thus half or three-~~thirds~~ of each color channel consist of ~~gaps~~ (black, unexposed photosites, since there is for example only one red-filtered photosite out of every four).		# and thus half (green) or three-fourths (red, blue) of each color channel consist of a lack of data (black, unexposed photosites, since there is for example only one red-filtered photosite out of every four).

	Displaying an image from a camera with a Bayer or X-Trans sensor is therefore not straight-forward - the mosaic of discrete data points need to be converted into a cogent color image. This process is called '''demosaicing'''.		Displaying an image from a camera with a Bayer or X-Trans sensor is therefore not straight-forward - the mosaic of discrete data points need to be converted into a cogent color image. This process is called '''demosaicing'''.

DrSlony: Rewrote most of the article, added Dual Demosaic #4961

2019-05-06T13:15:51Z

Rewrote most of the article, added Dual Demosaic #4961

Show changes

DrSlony at 20:49, 18 March 2019

2019-03-18T20:49:01Z

← Older revision		Revision as of 20:49, 18 March 2019
Line 53:		Line 53:
	Which demosaicing method should you use?<br>		Which demosaicing method should you use?<br>
	This page aims to tell you as much about the various algorithms as is relevant to a photographer, but there is no much to say as the explanations would quickly become technical and of a programmatical and mathematical nature. After reading though this article you should know that LMMSE and IGV are to be used on high ISO photos and that for the majority of other cases you should stick with the default AMaZE method, but of course you are free to explore for yourself and to test each method out on your own raw files. It is of no use reading an article somewhere on the internet which claims that some specific method is best because in their test that method was the sharpest, as the performance of each method depends specifically on the sensor in your camera and even on ISO, so keeping our suggestions in mind, run your own test and make up your own mind!		This page aims to tell you as much about the various algorithms as is relevant to a photographer, but there is no much to say as the explanations would quickly become technical and of a programmatical and mathematical nature. After reading though this article you should know that LMMSE and IGV are to be used on high ISO photos and that for the majority of other cases you should stick with the default AMaZE method, but of course you are free to explore for yourself and to test each method out on your own raw files. It is of no use reading an article somewhere on the internet which claims that some specific method is best because in their test that method was the sharpest, as the performance of each method depends specifically on the sensor in your camera and even on ISO, so keeping our suggestions in mind, run your own test and make up your own mind!

	~~RawTherapee versions 4.2.91 and newer always use the demosaicing method you choose regardless of zoom level.~~
	RawTherapee versions older than 4.2.91 use the Fast algorithm to initially open the image for editing. After this, the selected demosaicing method is applied when the image is zoomed to 100% magnification or when the detail window is opened. The selected method is also used for batch processing. It is not recommended to select the Fast method for the final conversion, as it is a low-quality algorithm for display purposes.

	'''AMaZE''' (Aliasing Minimization and Zipper Elimination) is the default demosaicing method, as it yields the best results in most cases. In RawTherapee versions 2.4 and older VNG4 used to be the preferred algorithm for Olympus cameras, as AMaZE didn't exist yet and VNG4 eliminated certain maze pattern artifacts that might have been created by the other methods, but with the introduction of the AMaZE method in RawTherapee version 3.0 Olympus users might prefer that instead.		'''AMaZE''' (Aliasing Minimization and Zipper Elimination) is the default demosaicing method, as it yields the best results in most cases. In RawTherapee versions 2.4 and older VNG4 used to be the preferred algorithm for Olympus cameras, as AMaZE didn't exist yet and VNG4 eliminated certain maze pattern artifacts that might have been created by the other methods, but with the introduction of the AMaZE method in RawTherapee version 3.0 Olympus users might prefer that instead.
Line 63:		Line 60:
	'''DCB''' produces similar results to AMaZE. AMaZE can often be slightly superior in recovering details, while DCB can be better at avoiding false colors especially in images from cameras without anti-aliasing filters.		'''DCB''' produces similar results to AMaZE. AMaZE can often be slightly superior in recovering details, while DCB can be better at avoiding false colors especially in images from cameras without anti-aliasing filters.

	~~When~~ working with very noisy, high ISO images in conjunction with the [[Noise Reduction]] tool~~, it is recommended to use the '''LMMSE''' or '''IGV''' demosaicing methods~~. They will prevent false maze patterns from appearing, and prevent the image from looking washed-out due to heavy noise reduction.		'''LMMSE''' and '''IGV''' are recommended when working with very noisy, high ISO images, in conjunction with the [[Noise Reduction]] tool. They will prevent false maze patterns from appearing, and prevent the image from looking washed-out due to heavy noise reduction. IGV is also quite effective at mitigating moiré patterns.

	If you use a medium format technical camera with near-symmetrical wide angle lenses such as the Schneider Digitar 28mm or 35mm it's likely that your file will contain some crosstalk, especially if the lens is shifted (due to the low angle of incoming light from these lenses some light leaks over to the next pixel on the sensor), and in this case you can get mazing artifacts with AMaZE and DCB because of green channel separation caused by the crosstalk. If you, via adapters, combine a mirrorless camera with a wide angle lens designed for film, you may also get crosstalk. It can then be better to use the more robust '''VNG4''' algorithm (Variable Number of Gradients) which handles this situation well, at the cost of some fine detail. An alternative is to enable [[Preprocessing#Green Equilibration\|green equilibration]] to even-out the green channel differences.		If you use a medium format technical camera with near-symmetrical wide angle lenses such as the Schneider Digitar 28mm or 35mm it's likely that your file will contain some crosstalk, especially if the lens is shifted (due to the low angle of incoming light from these lenses some light leaks over to the next pixel on the sensor), and in this case you can get mazing artifacts with AMaZE and DCB because of green channel separation caused by the crosstalk. If you, via adapters, combine a mirrorless camera with a wide angle lens designed for film, you may also get crosstalk. It can then be better to use the more robust '''VNG4''' algorithm (Variable Number of Gradients) which handles this situation well, at the cost of some fine detail. An alternative is to enable [[Preprocessing#Green Equilibration\|green equilibration]] to even-out the green channel differences.

Ingo: /* Method */

2018-11-29T19:04:00Z

Method

← Older revision		Revision as of 19:04, 29 November 2018
Line 58:		Line 58:

	'''AMaZE''' (Aliasing Minimization and Zipper Elimination) is the default demosaicing method, as it yields the best results in most cases. In RawTherapee versions 2.4 and older VNG4 used to be the preferred algorithm for Olympus cameras, as AMaZE didn't exist yet and VNG4 eliminated certain maze pattern artifacts that might have been created by the other methods, but with the introduction of the AMaZE method in RawTherapee version 3.0 Olympus users might prefer that instead.		'''AMaZE''' (Aliasing Minimization and Zipper Elimination) is the default demosaicing method, as it yields the best results in most cases. In RawTherapee versions 2.4 and older VNG4 used to be the preferred algorithm for Olympus cameras, as AMaZE didn't exist yet and VNG4 eliminated certain maze pattern artifacts that might have been created by the other methods, but with the introduction of the AMaZE method in RawTherapee version 3.0 Olympus users might prefer that instead.

			'''RCD''' does an excellent job for round edges, for example stars in astrophotography, while preserving almost the same level of detail as AMaZE.

	'''DCB''' produces similar results to AMaZE. AMaZE can often be slightly superior in recovering details, while DCB can be better at avoiding false colors especially in images from cameras without anti-aliasing filters.		'''DCB''' produces similar results to AMaZE. AMaZE can often be slightly superior in recovering details, while DCB can be better at avoiding false colors especially in images from cameras without anti-aliasing filters.

Ingo: /* Method */

2018-11-29T18:44:47Z

Method

← Older revision		Revision as of 18:44, 29 November 2018
Line 11:		Line 11:
	The following demosaicing algorithms are available for raw files from Bayer sensors:		The following demosaicing algorithms are available for raw files from Bayer sensors:
	* AMaZE		* AMaZE
			* AMaZE+VNG4
			* RCD
			* RCD+VNG4
			* DCB
			* DCB+VNG4
			* LMMSE
	* IGV		* IGV
	* ~~LMMSE~~		* AHD
	* EAHD		* EAHD
	* HPHD		* HPHD
	* VNG4		* VNG4
	* DCB
	* AHD
	* Fast		* Fast
	* Mono		* Mono
			* Pixel Shift
	* None		* None

	The following demosaicing algorithms are available for raw files from X-Trans sensors:		The following demosaicing algorithms are available for raw files from X-Trans sensors:
	* 3-Pass		* 3-pass+fast
	* 1-Pass		* 3-pass (Markesteijn)
			* 1-Pass+fast
			* 1-Pass (Markesteijn)
	* Fast		* Fast
	* Mono		* Mono

DrSlony: Added link to Green Equilibration

2017-08-23T07:37:12Z

Added link to Green Equilibration

← Older revision		Revision as of 07:37, 23 August 2017
Line 56:		Line 56:
	When working with very noisy, high ISO images in conjunction with the [[Noise Reduction]] tool, it is recommended to use the '''LMMSE''' or '''IGV''' demosaicing methods. They will prevent false maze patterns from appearing, and prevent the image from looking washed-out due to heavy noise reduction.		When working with very noisy, high ISO images in conjunction with the [[Noise Reduction]] tool, it is recommended to use the '''LMMSE''' or '''IGV''' demosaicing methods. They will prevent false maze patterns from appearing, and prevent the image from looking washed-out due to heavy noise reduction.

	If you use a medium format technical camera with near-symmetrical wide angle lenses such as the Schneider Digitar 28mm or 35mm it's likely that your file will contain some crosstalk, especially if the lens is shifted (due to the low angle of incoming light from these lenses some light leaks over to the next pixel on the sensor), and in this case you can get mazing artifacts with AMaZE and DCB because of green channel separation caused by the crosstalk. If you, via adapters, combine a mirrorless camera with a wide angle lens designed for film, you may also get crosstalk. It can then be better to use the more robust '''VNG4''' algorithm (Variable Number of Gradients) which handles this situation well, at the cost of some fine detail. An alternative is to enable green equilibration to even-out the green channel differences.		If you use a medium format technical camera with near-symmetrical wide angle lenses such as the Schneider Digitar 28mm or 35mm it's likely that your file will contain some crosstalk, especially if the lens is shifted (due to the low angle of incoming light from these lenses some light leaks over to the next pixel on the sensor), and in this case you can get mazing artifacts with AMaZE and DCB because of green channel separation caused by the crosstalk. If you, via adapters, combine a mirrorless camera with a wide angle lens designed for film, you may also get crosstalk. It can then be better to use the more robust '''VNG4''' algorithm (Variable Number of Gradients) which handles this situation well, at the cost of some fine detail. An alternative is to enable [[Preprocessing#Green Equilibration\|green equilibration]] to even-out the green channel differences.

	'''AHD''' (Adaptive Homogeneity-Directed), '''EAHD''' (Horvath's AHD) and '''HPHD''' (Heterogeneity-Projection Hard-Decision) are old methods which are generally slow and inferior to the other methods.		'''AHD''' (Adaptive Homogeneity-Directed), '''EAHD''' (Horvath's AHD) and '''HPHD''' (Heterogeneity-Projection Hard-Decision) are old methods which are generally slow and inferior to the other methods.

DrSlony: Reverted edits by DrSlony (talk) to last revision by Ingo

2017-08-15T08:48:28Z

Reverted edits by DrSlony (talk) to last revision by Ingo

← Older revision		Revision as of 08:48, 15 August 2017
Line 2:		Line 2:

	[[File:Chipincamera.jpg\|thumb\|Cutaway illustration of a camera showing the light sensor with a Bayer filter.]]		[[File:Chipincamera.jpg\|thumb\|Cutaway illustration of a camera showing the light sensor with a Bayer filter.]]
	[[~~Image~~:Bayer pattern on sensor.svg\|Bayer pattern on sensor.]]		[[File:Bayer pattern on sensor.svg\|thumb\|Bayer pattern on sensor.]]
	[[~~Image~~:Bayer pattern on sensor profile.svg\|thumb\|Profile/cross-section of sensor.]]		[[File:Bayer pattern on sensor profile.svg\|thumb\|Profile/cross-section of sensor.]]
	Most digital cameras today use a [https://en.wikipedia.org/wiki/Color_filter_array color filter array] over their sensors. To display such raw files their data needs to be [http://www.cambridgeincolour.com/tutorials/camera-sensors.htm demosaiced]. Cameras with a [https://en.wikipedia.org/wiki/Foveon_X3_sensor Foveon X3 sensor] (Sigma) do not have color filter arrays and so do not need demosaicing. Demosaicing is why opening a raw file always takes a bit longer than opening a [https://en.wikipedia.org/wiki/JPEG JPEG] or [https://en.wikipedia.org/wiki/TIFF TIFF] file, where the data is already 'display-ready'. RawTherapee offers several demosaicing algorithms, each with its own characteristics. The differences between them are often very subtle - one might need to zoom in to 200-400% to discern them - but since the program works on a pixel-by-pixel basis and demosaicing is the basis upon which all other tools work, the choice of demosaicing algorithm can have a visually significant effect when combined with other tools, such as the sharpening ones. The choice of a certain algorithm influences, among other things, the quality of very fine details in the image, whether false maze patterns will appear, and decides how well colored edges are rendered.		Most digital cameras today use a [https://en.wikipedia.org/wiki/Color_filter_array color filter array] over their sensors. To display such raw files their data needs to be [http://www.cambridgeincolour.com/tutorials/camera-sensors.htm demosaiced]. Cameras with a [https://en.wikipedia.org/wiki/Foveon_X3_sensor Foveon X3 sensor] (Sigma) do not have color filter arrays and so do not need demosaicing. Demosaicing is why opening a raw file always takes a bit longer than opening a [https://en.wikipedia.org/wiki/JPEG JPEG] or [https://en.wikipedia.org/wiki/TIFF TIFF] file, where the data is already 'display-ready'. RawTherapee offers several demosaicing algorithms, each with its own characteristics. The differences between them are often very subtle - one might need to zoom in to 200-400% to discern them - but since the program works on a pixel-by-pixel basis and demosaicing is the basis upon which all other tools work, the choice of demosaicing algorithm can have a visually significant effect when combined with other tools, such as the sharpening ones. The choice of a certain algorithm influences, among other things, the quality of very fine details in the image, whether false maze patterns will appear, and decides how well colored edges are rendered.