CN102722868A - Tone mapping method for high dynamic range image - Google Patents

Abstract

The invention discloses a tone mapping method for a high dynamic range image. The method comprises the following steps of: firstly, reading the high dynamic range image, and acquiring a log-domain brightness image of the high dynamic range image; secondly, estimating illumination components of the high dynamic range image by multi-scale weighted least squares (WLS) filtering to avoid the phenomenon of halo generated at a high light and shade contrast place; thirdly, recovering brightness components of the high dynamic range image by using a multi-scale Retinex method, eliminating the gray of a 1% part of each of two poles to prevent under-exposure and over-exposure, and adjusting the brightness degree of the image according to a self-adaptive adjustment function; and finally, recovering the color of the high dynamic range image by introducing an index control saturation method. The tone mapping method for the high dynamic range image is applicable to display and output of the high dynamic range image on low dynamic range equipment, generation of the halo can be avoided, and the displayed image has the characteristics of high contrast and good detail reproduction effect.

Description

A kind of high dynamic range image tone mapping method

Technical field

The invention belongs to the digital image processing techniques field, be used on the low-dynamic range display device, reproducing high dynamic range images, be specifically related to a kind of high dynamic range image tone mapping method.

Background technology

The brightness range of high dynamic range images is generally at 0.001cd/m ²To 100000cd/m ²Between.Because high dynamic range images can be good at reproducing gray scale abundant in the actual scene, produces effect true to nature, it is applied to fields such as video display, computer game, military affairs, astronomy and medical science detection just more and more widely.With respect to the brightness range of high dynamic range images, the brightness range that many display devices (like CRT, LCD, projector etc.) can produce then seems too narrow, and the brightness range that produces like common CRT approximately is 1cd/m ²To 100cd/m ²So, use regular display to show that the unmatched problem of dynamic range can appear in high dynamic range images.Therefore, need to adopt certain mode to compress the dynamic range of high dynamic range images, make it to mate output on the limited media of dynamic range, this process is called tone map (tone mapping).

Chinese scholars has proposed many tone mapping methods in recent years, mainly can be divided into two types: the 1) overall situation (Global) operator; 2) local (Local) operator.Overall situation operator only depends on the brightness value of pixel, and local operator not only relies on the brightness value of pixel, but also relevant with locations of pixels and neighborhood territory pixel brightness value.Local operator has been owing to considered the spatial neighborhood Pixel Information can keep more images detailed information and local contrast, thereby can obtain and shine upon effect preferably, becomes the emphasis of present concern.But local operator is prone to produce " halation (Halo) " phenomenon when handling high dynamic range images, influences display effect.

Summary of the invention

The purpose of this invention is to provide a kind of high dynamic range image tone mapping method, solved " halation " and the imperfect problem of display effect that exist in the existing method.

The technical scheme that the present invention adopts does, a kind of high dynamic range image tone mapping method is implemented according to following steps:

Step 1 is obtained the luminance picture L of high dynamic range images _In, and the formation log-domain luminance picture L that takes the logarithm, R wherein, G, B are respectively the red, green, blue color value of each pixel of high dynamic range images, and it is calculated as;

L _in＝0.299R+0.587G+0.114B，

L＝log(L _in+0.00001)；

Step 2 is carried out multiple dimensioned WLS filtering to L, estimates multiple dimensioned illumination component L _k, k=1 wherein, 2 ..., with " halation " phenomenon of avoiding the strong place of light and shade contrast to produce, it is expressed as:

L _k＝WLS(L,λ _k)，

Wherein operator WLS () representes to carry out WLS filtering, λ _kRepresent scale factor, its value has determined illumination component image L _kLevel and smooth degree, increase λ _k, will make level and smooth degree increase, can confirm its value according to the experiment situation usually;

Step 3 adopts multiple dimensioned Retinex method to recover the brightness composition of high dynamic range images, that is:

Wherein N represents yardstick number, ω _kBe the weight coefficient of different scale, and

Step 4 is to R _LGet index, obtain R ₁, that is: R ₁=exp (R _L);

Step 5, cancellation R ₁Two extreme 1% gray level prevents under-exposure and generation overexposure, and carries out normalization, obtains R ₂, that is:

$R_2=\left\{\begin{array}{cc}0& R_1<m_s\\ \frac{R_1-m_s}{m_l-m_s}& m_s\&le;R_1<m_l\\ 1& R_1\&GreaterEqual;m_l\end{array}\right.,$

M wherein _sBe R ₁After middle pixel value sorts from small to large, the R that 1% pixel place is corresponding ₁Value; m _lBe R ₁After middle pixel value sorts from small to large, the R that 99% pixel place is corresponding ₁Value;

Step 6 adopts the self-adaptation adjustment function to R ₂Regulate, and its gray-scale value is mapped to the scope of [0,255], that is:

Wherein, regulate the image that parameter b can obtain different bright-dark degrees, the span of b is [0.5,1.0];

Step 7 is introduced index control saturation degree, adopts following computing formula, carries out color recovery, promptly accomplishes the high dynamic range image tone mapping:

$\left\{\begin{array}{c}{R}_{\mathrm{out}}={\left(\frac{R}{L_{\mathrm{in}}}\right)}^s\&times;L_{\mathrm{out}}\\ G_{\mathrm{out}}={\left(\frac{G}{L_{\mathrm{in}}}\right)}^s\&times;L_{\mathrm{out}}\\ B_{\mathrm{out}}={\left(\frac{B}{L_{\mathrm{in}}}\right)}^s\&times;L_{\mathrm{out}}\end{array}\right..$

The invention has the beneficial effects as follows, adopt multiple dimensioned WLS wave filter to estimate the illumination component of high dynamic range images, overcome " halation " problem that exists in the tone map process preferably; Through removing two extreme 1% gray level and adopting the self-adaptation adjustment function, make image after the tone map have contrast height, characteristics that the details display effect is good, can on the low-dynamic range display device, can reproduce high dynamic range images better.

Description of drawings

Fig. 1 is embodiment 1 high dynamic range images that demonstrates after adopting the inventive method to handle;

Fig. 2 is the high dynamic range images that demonstrates after adopting existing high dynamic range image tone mapping method based on Retinex to handle;

Fig. 3 is that comparison diagram is amplified in the part at above-mentioned Fig. 1 black surround position and Fig. 2 black surround position.

Embodiment

Through embodiment the present invention is elaborated below.

High dynamic range image tone mapping method of the present invention, implement according to following steps:

Step 1 is obtained the luminance picture L of high dynamic range images _In, and the formation log-domain luminance picture L that takes the logarithm, R wherein, G, B are respectively the red, green, blue color value of each pixel of high dynamic range images, and it is calculated as;

L _in＝0.299R+0.587G+0.114B，

L＝log(L _in+0.00001)；

Step 2 is carried out multiple dimensioned WLS filtering to L, estimates multiple dimensioned illumination component L _k, k=1 wherein, 2 ..., with " halation " phenomenon of avoiding the strong place of light and shade contrast to produce, it is expressed as:

L _k＝WLS(L,λ _k)，

Wherein operator WLS () representes to carry out WLS filtering, λ _kRepresent scale factor, its value has determined illumination component image L _kLevel and smooth degree, increase λ _k, will make level and smooth degree increase, confirm its value according to the experiment situation usually.

For multiple dimensioned WLS filtering, it specifically is calculated as:

L _k＝(I+λ _kL _g) ^-1L，

Wherein, I is a unit matrix, λ _kBe scale factor, L _gFor:

$L_g=D_x^T{A}_x{D}_x+D_y^T{A}_y{D}_y,$

D wherein _x, D _yBe respectively the level and the vertical direction difference factor, T represents transpose of a matrix, A _x, A _yBe diagonal matrix, its diagonal entry is respectively a _{X, p}(L), a _{Y, p}(L), it is calculated as:

$\left\{\begin{array}{c}{a}_{x,p}\left(L\right)={({\left|\frac{\&PartialD;l}{\&PartialD;x}\left(p\right)\right|}^{\mathrm{\&alpha;}}+\mathrm{\&epsiv;})}^{-1}\\ a_{y,p}\left(L\right)={({\left|\frac{\&PartialD;l}{\&PartialD;y}\left(p\right)\right|}^{\mathrm{\&alpha;}}+\mathrm{\&epsiv;})}^{-1}\end{array}\right.,$

Wherein parameter alpha is used for representing to control the sensitivity of smoothing process to the image border, and span is 1.0-1.8; Parameter l=log (L+0.000001), constant ε value is 0.00001, mainly is in order to prevent that denominator from being zero situation appearance.

Step 3 adopts multiple dimensioned Retinex method to recover the brightness composition of high dynamic range images, that is:

N represents scale parameter, generally gets 3 yardsticks.

ω wherein _kBe the weight coefficient of different scale, and

Step 4, the R that step 3 is obtained _LGet index, obtain R ₁, that is: R ₁=exp (R _L);

Step 5, cancellation R ₁Two extreme 1% gray level prevents under-exposure and generation overexposure, and carries out normalization, obtains R ₂, that is:

$R_2=\left\{\begin{array}{cc}0& R_1<m_s\\ \frac{R_1-m_s}{m_l-m_s}& m_s\&le;R_1<m_l\\ 1& R_1\&GreaterEqual;m_l\end{array}\right.,$

Wherein, m _sBe R ₁After middle pixel value sorts from small to large, the R that 1% pixel place is corresponding ₁Value; m _lBe R ₁After middle pixel value sorts from small to large, the R that 99% pixel place is corresponding ₁Value;

Step 6 adopts the self-adaptation adjustment function to R ₂Regulate, and its gray-scale value is mapped to the scope of [0,255], that is:

Wherein, the numerical value of regulating parameter b can obtain the image of different bright-dark degrees, and the span of b is [0.5,1.0];

Step 7 is introduced index control saturation degree, adopts following computing formula, carries out color recovery:

$\left\{\begin{array}{c}{R}_{\mathrm{Out}}={\left(\frac{R}{L_{\mathrm{In}}}\right)}^s\&times;L_{\mathrm{Out}}\\ G_{\mathrm{Out}}={\left(\frac{G}{L_{\mathrm{In}}}\right)}^s\&times;L_{\mathrm{Out}}\\ B_{\mathrm{Out}}={\left(\frac{B}{L_{\mathrm{In}}}\right)}^s\&times;L_{\mathrm{Out}}\end{array}\right..$ Promptly accomplish the high dynamic range image tone mapping.

Embodiment 1

High dynamic range image tone mapping method of the present invention, implement through following steps and concrete parameter:

Step 1 reads high dynamic range images, is the data of the rgb format of integer with the data-switching of its floating type, calculates the luminance picture L of this high dynamic range images then _In, and the formation log-domain luminance picture L that takes the logarithm, it is calculated as;

L _in＝0.299R+0.587G+0.114B，

L＝log(L _in+0.00001)，

Step 2 is carried out multiple dimensioned WLS filtering to L, estimates multiple dimensioned illumination component L _k(k=1,2, L), with " halation " phenomenon of avoiding the strong place of light and shade contrast to produce.It can be expressed as:

L _k＝WLS(L,λ _k)，

Wherein operator WLS () representes to carry out WLS filtering, λ _kRepresent scale factor, its value has determined illumination component image L _kLevel and smooth degree, increase λ _k, will make level and smooth degree increase, confirm its value according to the experiment situation usually.In the present embodiment, adopt three yardsticks, i.e. k=1,2,3, get λ ₁=1, λ ₂=5, λ ₃=25.

For multiple dimensioned WLS filtering, it specifically is calculated as: L _k=(I+ λ _kL _g) ^-1L,

Wherein I is a unit matrix, λ _kBe scale factor, L _gFor:

Wherein, D _x, D _yBe respectively the level and the vertical direction difference factor, A _x, A _yBe diagonal matrix, its diagonal entry is respectively a _{X, p}(L), a _{Y, p}(L), it is calculated as:

$\left\{\begin{array}{c}{a}_{x,p}\left(L\right)={({\left|\frac{\&PartialD;l}{\&PartialD;x}\left(p\right)\right|}^{\mathrm{\&alpha;}}+\mathrm{\&epsiv;})}^{-1}\\ a_{y,p}\left(L\right)={({\left|\frac{\&PartialD;l}{\&PartialD;y}\left(p\right)\right|}^{\mathrm{\&alpha;}}+\mathrm{\&epsiv;})}^{-1}\end{array}\right.,$ Wherein parameter alpha is got α=1.5 in the present embodiment.

Step 3 adopts multiple dimensioned Retinex method to recover the brightness composition of high dynamic range images, that is:

ω wherein _kBe the weight coefficient of different scale, and

In the present embodiment, get 3 yardsticks, get ω accordingly ₁=0.25, ω ₂=0.5, ω ₃=0.25.Take

Step 4 is to R _LGet index, obtain R ₁, that is: R ₁=exp (R _L);

Step 5, cancellation R ₁Two extreme 1% gray level prevents under-exposure and generation overexposure, and carries out normalization, obtains R ₂, that is:

$R_2=\left\{\begin{array}{cc}0& R_1<m_s\\ \frac{R_1-m_s}{m_l-m_s}& m_s\&le;R_1<m_l\\ 1& R_1\&GreaterEqual;m_l\end{array}\right.,$

M wherein _sBe R ₁After middle pixel value sorts from small to large, the R that 1% pixel place is corresponding ₁Value; m _lBe R ₁After middle pixel value sorts from small to large, the R that 99% pixel place is corresponding ₁Value;

Step 6 adopts the self-adaptation adjustment function to R ₂Regulate, and its gray-scale value is mapped to the scope of [0,255], that is:

Regulate parameter b in the present embodiment, get b=0.78.

Step 7 is introduced index control saturation degree, adopts following computing formula, in the present embodiment, gets s=12.2.

Carry out color recovery, can accomplish the high dynamic range image tone mapping.

$\left\{\begin{array}{c}{R}_{\mathrm{out}}={\left(\frac{R}{L_{\mathrm{in}}}\right)}^s\&times;L_{\mathrm{out}}\\ G_{\mathrm{out}}={\left(\frac{G}{L_{\mathrm{in}}}\right)}^s\&times;L_{\mathrm{out}}\\ B_{\mathrm{out}}={\left(\frac{B}{L_{\mathrm{in}}}\right)}^s\&times;L_{\mathrm{out}}\end{array}\right..$

Fig. 1 is that the embodiment of the invention 1 is handled the high dynamic range images that the back shows.For the effect of the inventive method is described better; Fig. 2 adopts existing method; Promptly handle the high dynamic range images that the back shows based on the high dynamic range image tone mapping method of Retinex; Fig. 3 is that comparison diagram is amplified in the part of Fig. 1 and Fig. 2, and wherein the first half of Fig. 3 is the partial enlarged drawing at the black surround position among Fig. 1, and the Lower Half of Fig. 3 is the partial enlarged drawing at the black surround position among Fig. 2.Can contrast from Fig. 3 and to find out, adopt the inventive method to compare, can remove " halation " phenomenon that high dynamic range image tone mapping back exists better with existing method based on Retinex.What be worth explanation is, the inventive method and existing method are colored images to what handle, limit owing to declare the requirement of text, and institute's drawings attached has become gray-scale map and shown, its " halation " phenomenon does not have coloured image true to nature.

High dynamic range image tone mapping method of the present invention, to " halation " phenomenon that is prone in the localization tone mapping method produce, through WLS Filtering Estimation illumination component, " halation " phenomenon that can avoid the strong place of light and shade contrast to produce; Through removing the gray level at two extreme 1% places, prevent under-exposure and generation overexposure, adopt the self-adaptation adjustment function that the light and shade of image is regulated, make image after the tone map have contrast height, characteristics that the details display effect is good.

Claims (2)

1. a high dynamic range image tone mapping method is characterized in that, implements according to following steps:

Step 1 is obtained the luminance picture L of high dynamic range images _In, and the formation log-domain luminance picture L that takes the logarithm, R wherein, G, B are respectively the red, green, blue color value of each pixel of high dynamic range images, and it is calculated as;

L _in＝0.299R+0.587G+0.114B，

L=log(L _in+0.00001)；

Step 2 is carried out multiple dimensioned WLS filtering to L, estimates multiple dimensioned illumination component L _k, k=1 wherein, 2 ..., with " halation " phenomenon of avoiding the strong place of light and shade contrast to produce, it is expressed as:

L _k＝WLS(L,λ _k)，

Wherein operator WLS () representes to carry out WLS filtering, λ _kRepresent scale factor, its value has determined illumination component image L _kLevel and smooth degree, increase λ _k, will make level and smooth degree increase;

Step 3 adopts multiple dimensioned Retinex method to recover the brightness composition of high dynamic range images, that is:

Wherein N represents yardstick number, ω _kBe the weight coefficient of different scale, and

Step 4 is to R _LGet index, obtain R ₁, that is: R ₁=exp (R _L);

Step 5, cancellation R ₁Two extreme 1% gray level prevents under-exposure and generation overexposure, and carries out normalization, obtains R ₂, that is:

$R_2=\left\{\begin{array}{cc}0& R_1<m_s\\ \frac{R_1-m_s}{m_l-m_s}& m_s\&le;R_1<m_l\\ 1& R_1\&GreaterEqual;m_l\end{array}\right.,$

Wherein, m _sBe R ₁After middle pixel value sorts from small to large, the R that 1% pixel place is corresponding ₁Value; m _lBe R ₁After middle pixel value sorts from small to large, the R that 99% pixel place is corresponding ₁Value;

Step 6 adopts the self-adaptation adjustment function to R ₂Regulate, and its gray-scale value is mapped to the scope of [0,255], that is:

Wherein, regulate the image that parameter b can obtain different bright-dark degrees, the span of b is [0.5,1.0];

Step 7 is introduced index control saturation degree, adopts following computing formula, carries out color recovery:

$\left\{\begin{array}{c}{R}_{\mathrm{Out}}={\left(\frac{R}{L_{\mathrm{In}}}\right)}^s\&times;L_{\mathrm{Out}}\\ G_{\mathrm{Out}}={\left(\frac{G}{L_{\mathrm{In}}}\right)}^s\&times;L_{\mathrm{Out}}\\ B_{\mathrm{Out}}={\left(\frac{B}{L_{\mathrm{In}}}\right)}^s\&times;L_{\mathrm{Out}}\end{array}\right.,$ Promptly accomplish the high dynamic range image tone mapping.

2. a kind of high dynamic range image tone mapping method according to claim 1 is characterized in that:

Multiple dimensioned WLS in the said step 2 is filtered into: L _k=(I+ λ _kL _g) ^-1L,

Wherein, I is a unit matrix, λ _kBe scale factor, L _gFor:

Wherein, D _x, D _yBe respectively the level and the vertical direction difference factor, T represents transpose of a matrix, A _x, A _yBe diagonal matrix, its diagonal entry is respectively a _{X, p}(L), a _{Y, p}(L), it is calculated as:

$\left\{\begin{array}{c}{a}_{x,p}\left(L\right)={({\left|\frac{\&PartialD;l}{\&PartialD;x}\left(p\right)\right|}^{\mathrm{\&alpha;}}+\mathrm{\&epsiv;})}^{-1}\\ a_{y,p}\left(L\right)={({\left|\frac{\&PartialD;l}{\&PartialD;y}\left(p\right)\right|}^{\mathrm{\&alpha;}}+\mathrm{\&epsiv;})}^{-1}\end{array}\right.,$

Wherein parameter alpha is used for controlling the sensitivity of smoothing process to the image border, and span is 1.0-1.8;

Parameter l=log (L+0.000001), constant ε are 0.00001.

Images