CN101212545A - Method and device for mapping high-dynamic range graphics to low-dynamic range graphics - Google Patents

Abstract

The invention discloses a method for realizing a reflection of graphics from high dynamic range to low dynamic range and a device thereof, which relates to a graph processing technique, providing a method for compressing a reflection from high dynamic range to low dynamic range and a device thereof which can effectively guarantee real-time requests and protect more details. The method comprises the steps: firstly the RGB value of every pixel is read from graphics texture of high dynamic range; the RGB value is transformed into practical light intensity value; then the practical light intensity value is transformed into average light intensity value; the light intensity value of every pixel is zoomed according to the necessary bright effect after reflection; at last the light intensity value of pixel which is reflected is calculated by the formula Ld(x, y)={L(x, y)[1+L(x, y)/L2white]+ Sigma }/1+L(x,y). The invention can compress the graphics of high dynamic range high-effectively and preserve the details and contrast, then reflect to the low dynamic range to suit the present regular display equipment in the low dynamic range. The invention also can effectively guarantee the real-time requests, and has the advantages of easiness in realizing and good scalability. The invention meets the requests of three-dimensional scenes (such as three-dimensional games) and virtual reality.

Description

A kind of figure of high dynamic range is mapped to the method and the device of low-dynamic range

Technical field

The present invention relates to a kind of graph processing technique, particularly a kind of figure compressing mapping of high dynamic range is to method and the device of low-dynamic range to adapt to the existing conventional display device.

Background technology

In recent years, (High Dynamic Range, HDR) technology began to have increasing application in fields such as computer graphics, virtual realities to high dynamic range.HDR is relative with traditional LDR (lowdynamic range), LDR adopts 8 texture formats (every pixel 24 or 32 colors), this brand-new model of HDR is represented the RGB and the brightness value of each pixel with actual physics parameter or linear function, parameter no longer is limited to integer, can reach bigger scope and pinpoint accuracy more.

The ratio of the maximum of piece image intensity level and minimum value is called as dynamic range (DynamicRange).By the automatic adjusting of eye pupil, to starlight, the dynamic range that human eye can be differentiated object can reach 100000000: 1 from bright daylight, even in same adaptation figure, need not regulate, human eye also can be differentiated 10000: 1 brightness range.Yet the luminance dynamic range that conventional display device can be rebuild only is 100: 1.

The figure of HDR is a large amount of existing in actual life, therefore needs in some way the dynamic range of image to be carried out convergent-divergent, makes it to mate the conventional display device that can only export LDR.This mode is called color range and rebuilds (Tone Reproduction) or tone mapping (Tone Mapping), it provide a kind of mode with real figure the brightness value convergent-divergent or be mapped to the scope that display device can show.Except the compression brightness range, also must keep the organoleptic quality (Perceptual Quality) of original image, must keep information such as original image contrast, bright degree, details as method for reconstructing.

The method for reconstructing of dynamic range roughly is divided into two kinds, and a kind of is spatial domain constant (Spatially Uniform), perhaps is the global scope dynamic compression.These class methods are when carrying out the dynamic range conversion to image, and each pixel (Pixel) goes up uses same conversion curve, and conversion curve can specify or obtain according to the content of image in advance.The dynamic range reconstruction technique of adjusting based on histogram with people (A visibility matching tone reproductionoperator for high dynamic range scenes) such as Ward Larson in this class algorithm is a sign, its deficiency is each zone that constant conversion curve can not adapting to image, causes result images to lose on details, color, bright degree.

Another kind is that the spatial domain changes (Spatially Varying), perhaps is local dynamic range compression.These class methods are carried out different conversion at the different zone of image.According to human visual system's (HVS) different models, various distinct methods are standard with what keep picture quality all in compression of dynamic range in a certain respect.Until the latter stage nineties, the whole bag of tricks all is to adjust at different HVS models on multilayered model, but because the filter function characteristic that adopts on the low-frequency image is not good, the serious halation (halo) that object edge produces in result images is these class methods of puzzlement problems for many years always.1999, people such as Tumblin have proposed LCIS method (Lcis:A boundary hierarchy for detail-preserving contrastreduction), by definition to the different details of image, improved resultant image quality, but make the operation of the method very poor efficiency that becomes, speed is slow excessively.

Goal of the invention

The objective of the invention is to solve the prior art above shortcomings, provide a kind of high dynamic range compression to be mapped to the method and apparatus of low-dynamic range, can ensure the real-time requirement effectively, efficient is higher, can also keep the details, color of figure, bright degree etc., and be easy to realize, be with good expansibility.

Purpose of the present invention realizes by following technical proposals:

A kind of figure of high dynamic range is mapped to the method for low-dynamic range, comprises the steps:

A, elder generation read the rgb value of each pixel from the graphical textures of high dynamic range;

B, convert rgb value to the actual light intensity value;

C, utilize following formula to convert the actual light intensity value to the average intensity value then,

${\stackrel{\_}{L}}_w=\exp \left[\frac{1}{N}\underset{x,y}{\mathrm{\Σ}}\log (\mathrm{\δ}+L_w(x,y))\right]$

Wherein, L _w(x, y) expression actual light intensity value;

X, y represent the two-dimensional coordinate of each pixel;

Expression average intensity value;

N represents the quantity of texture pixel;

δ is a very little constant, can be taken as 0.0001;

D, the bright effect required according to mapping back are come the light intensity value of each pixel of convergent-divergent by following formula,

$L(x,y)=\frac{a{L}_w(x,y)}{{\stackrel{\‾}{L}}_w}$

Wherein, the span of α for more gloomy figure, is got lower α value between 0 to 1, otherwise, then get higher α value;

E, at last by following formula, calculate the light intensity value of mapped pixel,

$L_d(x,y)=\frac{L(x,y)[1+\frac{L(x,y)}{L_{\mathrm{white}}^2}]+\mathrm{\σ}}{1+L(x,y)}$

Wherein, L _(x y) is the light intensity value of mapped pixel;

L _WhiteIt is the largest light intensity value;

σ is a self-defining value, when figure does not need complete black local time, can be set to nonzero value by σ, as 0.005.

Among the described step b, convert rgb value to actual light intensity value by following formula:

L _w＝0.27R+0.67G+0.06B

In the described steps d, the general recommendations of α value gets 0.18.

In above-mentioned each technical scheme, among the described step c, in frame buffer, read in the light intensity value of each pixel, in two pipelines, carry out, in first pipeline, create a little render target, draw a rectangular blocks and cover whole render target, on each pixel in target cache, pixel shader program add up light intensity value in the source cache and storage computation result; In second pipeline, the target cache in first pipeline that superposes is moved back-page calculating then, then includes in the texture of generation

A kind of figure of high dynamic range is mapped to the device of low-dynamic range, includes:

Pixel rgb value reader is used for reading from the graphical textures of high dynamic range the rgb value of each pixel;

Rgb value light intensity transducer is used for converting rgb value to the actual light intensity value;

Light intensity mean value transducer is used for the actual light intensity value is converted to the average intensity value by following formula,

${\stackrel{\_}{L}}_w=\exp \left[\frac{1}{N}\underset{x,y}{\mathrm{\Σ}}\log (\mathrm{\δ}+L_w(x,y))\right]$

Wherein, L _w(x, y) expression actual light intensity value;

X, y represent the two-dimensional coordinate of each pixel;

Expression average intensity value;

N represents the quantity of texture pixel;

δ is a very little constant, can be taken as 0.0001;

Pixel light intensity value scaler is used for coming the light intensity value of each pixel of convergent-divergent according to the required bright effect in mapping back by following formula,

$L(x,y)=\frac{a{L}_w(x,y)}{{\stackrel{\_}{L}}_w}$

Wherein, the span of α for more gloomy figure, is got lower α value between 0 to 1, otherwise, then get higher α value;

Pixel light intensity value mapper is used for obtaining by following formula the light intensity value of mapped pixel,

$L_d(x,y)=\frac{L(x,y)[1+\frac{L(x,y)}{L_{\mathrm{white}}^2}]+\mathrm{\σ}}{1+L(x,y)}$

Wherein, L _(x y) is the light intensity value of mapped pixel;

L _WhiteIt is the largest light intensity value;

σ is a self-defining value, when figure does not need complete black local time, can be set to nonzero value by σ, as be taken as 0.005.

Include the calculator that rgb value is converted to the actual light intensity value by following formula in the described rgb value light intensity transducer:

L _w＝0.27R+0.67G+0.06B

In the described pixel light intensity value scaler, the general recommendations of α value is taken as 0.18.

Described light intensity mean value transducer comprises frame buffer and two pipelines, read in the light intensity value that each pixel is arranged in the frame buffer, in first pipeline, create a little render target piece is arranged, be provided with a rectangular blocks and cover whole render target piece, also has the light intensity value accumulative register in first pipeline, utilize on pixel shader program each pixel in target cache the light intensity value in the source cache that adds up, and storage computation result; Include the stack calculator in second pipeline, the target cache of first pipeline that is used to superpose is moved back-page calculating then, then includes in the texture of generation

The present invention adopts said method or device, can compress the high dynamic range figure efficiently, and preserve its details and contrast, be mapped to low-dynamic range, the display device that can adapt to the existing conventional low-dynamic range, and can ensure the real-time requirement effectively, have be easy to realize, good advantages such as extensibility, can satisfy the requirement of three-dimensional scenic (as 3d gaming), virtual reality.

Embodiment

The present invention is further illustrated below in conjunction with specific embodiment.

Embodiment 1: a kind of figure of high dynamic range is mapped to the method for low-dynamic range, comprises the steps:

A, elder generation read the rgb value of each pixel from the graphical textures of high dynamic range;

B, convert rgb value to the actual light intensity value by following formula:

L _w＝0.27R+0.67G+0.06B；

C, utilize following formula to convert the actual light intensity value to the average intensity value then,

${\stackrel{\_}{L}}_w=\exp \left[\frac{1}{N}\underset{x,y}{\mathrm{\Σ}}\log (\mathrm{\δ}+L_w(x,y))\right]$

Wherein, L _w(x, y) expression actual light intensity value;

X, y represent the two-dimensional coordinate of each pixel;

Expression average intensity value;

N represents the quantity of texture pixel;

δ＝0.0001；

D, the bright effect required according to mapping back are come the light intensity value of each pixel of convergent-divergent by following formula,

$L(x,y)=\frac{a{L}_w(x,y)}{{\stackrel{\‾}{L}}_w}$

Wherein, the span of α for more gloomy figure, is got lower α value between 0 to 1, otherwise, then get higher α value, be taken as 0.18 in the present embodiment;

E, at last by following formula, calculate the light intensity value of mapped pixel,

$L_d(x,y)=\frac{L(x,y)[1+\frac{L(x,y)}{L_{\mathrm{white}}^2}]+\mathrm{\σ}}{1+L(x,y)}$

Wherein, L _(x y) is the light intensity value of mapped pixel, and this value promptly is the corresponding high dynamically value of light intensity that can show on conventional low-dynamic range display;

L _WhiteBe the largest light intensity value, this is a HDR value, and brightness value is greater than L _WhitePixel all can be mapped as complete white, L _WhiteCan be set as an infinitary value, be used for the light intensity value that produces is mapped as a value that can show;

σ is a self-defining value, when figure does not need complete black local time, can be set to nonzero value by σ,

Be taken as 0.005 in the present embodiment.

In the present embodiment, described step c carries out in two pipelines, in first pipeline, create a little render target, draw a rectangular blocks and cover whole render target, on each pixel in target cache, pixel shader program add up light intensity value in the source cache and storage computation result; In second pipeline, the target cache in first pipeline that superposes is moved back-page calculating then, then includes in the texture of generation

In the present embodiment, taked the frame per second of said method front and back to contrast as following table:

	Do not adopt this method	Adopt this method
			Current frame rate	136.727fps	77.8443fps

Average frame per second	136.938fps	77.8844fps
			The worst frame per second	122.388fps	75.0247fps
Best frame per second	137.177fps	137.117fps

Embodiment 2: basic step such as embodiment 1, wherein the α value is 0.42.

Embodiment 3: basic step such as embodiment 1, wherein the α value is 0.56.

Embodiment 4: basic step such as embodiment 1, wherein the α value is 0.78.

Embodiment 5: basic step such as embodiment 1, wherein the α value is 0.88.

Embodiment 6: a kind of figure of high dynamic range is mapped to the device of low-dynamic range, includes:

Pixel rgb value reader is used for reading from the graphical textures of high dynamic range the rgb value of each pixel;

Rgb value light intensity transducer is used for converting rgb value to the actual light intensity value by following formula:

L _w＝0.27R+0.67G+0.06B；

Light intensity mean value transducer is used for the actual light intensity value is converted to the average intensity value by following formula,

${\stackrel{\‾}{L}}_w=\exp \left[\frac{1}{N}\underset{x,y}{\mathrm{\Σ}}\log (\mathrm{\δ}+L_w(x,y))\right]$

Wherein, L _w(x, y) expression actual light intensity value;

X, y represent the two-dimensional coordinate of each pixel;

Expression average intensity value;

N represents the quantity of texture pixel;

δ＝0.0001；

Pixel light intensity value scaler is used for coming the light intensity value of each pixel of convergent-divergent according to the required bright effect in mapping back by following formula,

$L(x,y)=\frac{a{L}_w(x,y)}{{\stackrel{\‾}{L}}_w}$

Wherein, the span of α for more gloomy figure, is got lower α value between 0 to 1, otherwise, then get higher α value, be taken as 0.18 in the present embodiment;

Pixel light intensity value mapper is used for obtaining by following formula the light intensity value of mapped pixel,

$L_d(x,y)=\frac{L(x,y)[1+\frac{L(x,y)}{L_{\mathrm{white}}^2}]+\mathrm{\σ}}{1+L(x,y)}$

Wherein, L _(x y) is the light intensity value of mapped pixel, and this value promptly is the corresponding high dynamically value of light intensity that can show on conventional low-dynamic range display.；

L _WhiteBe the largest light intensity value, this is a HDR value, and brightness value is greater than L _WhitePixel all can be mapped as complete white, L _WhiteCan be set as an infinitary value, be used for the light intensity value that produces is mapped as a value that can show;

σ is a self-defining value, when figure does not need complete black local time, can be set to nonzero value by σ, is taken as 0.005 in the present embodiment.

In the present embodiment, light intensity mean value transducer comprises frame buffer and two pipelines, read in the light intensity value that each pixel is arranged in the frame buffer, in first pipeline, create a little render target piece is arranged, be provided with a rectangular blocks and cover whole render target piece, also have the light intensity value accumulative register in first pipeline, utilize on pixel shader program each pixel in target cache, light intensity value in the source cache and storage computation result add up; Include the stack calculator in second pipeline, the target cache of first pipeline that is used for superposeing is moved back-page calculating then, then includes in the texture of generation

Embodiment 7: basic structure such as embodiment 1, wherein the α value is 0.42.

Embodiment 8: basic structure such as embodiment 1, wherein the α value is 0.56.

Embodiment 9: basic structure such as embodiment 1, wherein the α value is 0.78.

Embodiment 10: basic structure such as embodiment 1, wherein the α value is 0.88.

Claims (8)

1. the figure of a high dynamic range is mapped to the method for low-dynamic range, it is characterized in that: comprise the steps,

A, elder generation read the rgb value of each pixel from the graphical textures of high dynamic range;

B, convert rgb value to the actual light intensity value;

C, utilize following formula to convert the actual light intensity value to the average intensity value then,

${\stackrel{\_}{L}}_w=\exp \left[\frac{1}{N}\underset{x,y}{\mathrm{\Σ}}\log (\mathrm{\δ}+L_w(x,y))\right]$

Wherein, L _w(x, y) expression actual light intensity value;

X, y represent the two-dimensional coordinate of each pixel;

Expression average intensity value;

N represents the quantity of texture pixel;

δ is a very little constant, can be taken as 0.0001;

D, the bright effect required according to mapping back are come the light intensity value of each pixel of convergent-divergent by following formula,

$L(x,y)=\frac{a{L}_w(x,y)}{{\stackrel{\‾}{L}}_w}$

Wherein, the span of α for more gloomy figure, is got lower α value between 0 to 1, otherwise, then get higher α value;

E, at last by following formula, calculate the light intensity value of mapped pixel,

$L_d(x,y)=\frac{L(x,y)[1+\frac{L(x,y)}{L_{\mathrm{white}}^2}]+\mathrm{\σ}}{1+L(x,y)}$

Wherein, L _(x y) is the light intensity value of mapped pixel;

L _WhiteIt is the largest light intensity value;

σ is a self-defining value, when figure does not need complete black local time, can be set to nonzero value by σ, as 0.005.

2. a kind of figure of high dynamic range is mapped to the method for low-dynamic range according to claim 1, it is characterized in that: among the described step b, convert rgb value to actual light intensity value by following formula,

L _w＝0.27R+0.67G+0.06B。

3. be mapped to the method for low-dynamic range as the figure of a kind of high dynamic range as described in the claim 2, it is characterized in that: in the described steps d, the α value gets 0.18.

4. be mapped to the method for low-dynamic range as the figure of a kind of high dynamic range as described in claim 1 or 2 or 3, it is characterized in that: among the described step c, in frame buffer, read in the light intensity value of each pixel, in two pipelines, carry out, in first pipeline, create a little render target, draw a rectangular blocks and cover whole render target, on each pixel in target cache, pixel shader program add up light intensity value in the source cache and storage computation result; In second pipeline, the target cache of first pipeline that superposes is moved back-page calculating then, then includes in the texture of generation

5. the figure of a high dynamic range is mapped to the device of low-dynamic range, it is characterized in that: include,

Pixel rgb value reader is used for reading from the graphical textures of high dynamic range the rgb value of each pixel;

Rgb value light intensity transducer is used for converting rgb value to the actual light intensity value;

Light intensity mean value transducer is used for the actual light intensity value is converted to the average intensity value by following formula,

${\stackrel{\_}{L}}_w=\exp \left[\frac{1}{N}\underset{x,y}{\mathrm{\Σ}}\log (\mathrm{\δ}+L_w(x,y))\right]$

Wherein, L _w(x, y) expression actual light intensity value;

X, y represent the two-dimensional coordinate of each pixel;

Expression average intensity value;

N represents the quantity of texture pixel;

δ is a very little constant, can be taken as 0.0001;

Pixel light intensity value scaler is used for coming the light intensity value of each pixel of convergent-divergent according to the required bright effect in mapping back by following formula,

$L(x,y)=\frac{a{L}_w(x,y)}{{\stackrel{\_}{L}}_w}$

Wherein, the span of α for more gloomy figure, is got lower α value between 0 to 1, otherwise, then get higher α value;

Pixel light intensity value mapper is used for obtaining by following formula the light intensity value of mapped pixel,

$L_d(x,y)=\frac{L(x,y)[1+\frac{L(x,y)}{L_{\mathrm{white}}^2}]+\mathrm{\σ}}{1+L(x,y)}$

Wherein, L _(x y) is the light intensity value of mapped pixel;

L _WhiteIt is the largest light intensity value;

σ is a self-defining value, when figure does not need complete black local time, can be set to nonzero value by σ, as 0.005.

6. be mapped to the device of low-dynamic range as the figure of a kind of high dynamic range as described in the claim 5, it is characterized in that: include in the described rgb value light intensity transducer by following formula and convert rgb value the calculator of actual light intensity value to,

L _w＝0.27R+0.67G+0.06B

7. be mapped to the device of low-dynamic range as the figure of a kind of high dynamic range as described in the claim 6, it is characterized in that: in the described pixel light intensity value scaler, the α value is taken as 0.18.

8. be mapped to the device of low-dynamic range as the figure of a kind of high dynamic range as described in claim 5 or 6 or 7, it is characterized in that: described light intensity mean value transducer comprises frame buffer and two pipelines, the frame buffer internal memory contains the light intensity value of each pixel, in first pipeline, create a little render target piece is arranged, be provided with a rectangular blocks and cover whole render target piece, also has the light intensity value accumulative register in first pipeline, utilize on pixel shader program each pixel in target cache, light intensity value in the source cache and storage computation result add up; Include the stack calculator in second pipeline, the target cache of first pipeline that is used for superposeing is moved back-page calculating then, then includes in the texture of generation