CN103020998A - Tone mapping method based on edge-preserving total variation model - Google Patents

Abstract

The invention discloses a tone mapping method based on an edge-preserving total variation model, which comprises the following steps of: inputting a high dynamic image; performing light intensity reconstruction on the input high dynamic image; obtaining a logarithm value of a reconstructed light intensity value; using an edge-preserving total variation filter to filter the obtained light intensity logarithm value to obtain a basic layer of the image; subtracting the basic layer of the image by the light intensity logarithm value to obtain a detail layer of the image; calculating a compression scaling factor and the output light intensity of the image; dividing the input light intensity of the image by the compressed light intensity to obtain a new scaling factor M; acting the scaling factor M respectively on each channel of a red channel, a green channel and a blue channel to obtain the image with compressed colors; and performing gamma calibration on the obtained image with the compressed colors to obtain a low dynamic image. The tone mapping method based on the edge-preserving total variation model can obtain the low dynamic image with very high quality, and does not produce halos.

Description

A kind of tone mapping method that keeps full sub-model based on the edge

Technical field

The present invention relates to technical field of image processing, be specifically related to keep based on the edge tone mapping method of full sub-model.

Background technology

High-dynamics image (High Dynamic Range Image, HDR Image) can be supported very large range of light intensities, can effectively store the intensity information of real world, presents the image of superior quality to people.Yet traditional display device and printer can not show these high-dynamics images.Because their displayable contrasts are far smaller than high-dynamics image.In order to address this problem, can to compress at present the high-dynamics image dynamic range and make it adapt to low dynamic image display device.Be exactly tone mapping method as this method that high-dynamics image is converted into low dynamic image (LowDynamic Range Image, LDR Image).

Now, a large amount of tone mapping methods is suggested.The simplest contrast compression be multiply by a scale factor C with original image, C＜1 exactly.This can lose details and the texture of shade in the image scene or high-brightness region.This just requires us to seek a kind of method to compress as far as possible contrast, keep simultaneously texture and details in the image.A lot of tone mapping methods have used the picture breakdown technology.Piece image is decomposed into basal layer and levels of detail.Basal layer has high-contrast to be needed compressedly, and levels of detail has clearly that texture need to be retained.Only basal layer is compressed, levels of detail remains unchanged, and is then that the basal layer after levels of detail and the compression is synthetic, just obtained low dynamic image.Basal layer can obtain by the edge preserving smoothing method, and levels of detail deducts basal layer by input picture and obtains.Durand and Dorsey document " Fast bilateralfiltering for the display of high-dynamic-range images " (2002, pp.257-266) introduced a kind of tone mapping method based on quick bilateral filtering.The quick two-sided filter of this utilization carries out multiple dimensioned decomposition to image, obtains basal layer and levels of detail.Quick two-sided filter effectively level and smooth small details keeps strong edge simultaneously.But, increasing smoothness, the image border will be blured.This is so that halation phenomenon appears in the LDR image that obtains by the method.

Based on the tone mapping method of quick bilateral filtering effectively level and smooth small details keep simultaneously strong edge.But, increasing smoothness, the image border will be blured.This is so that halation phenomenon appears in the LDR image that obtains by the method.

Summary of the invention

The present invention has overcome the shortcoming that can produce halation phenomenon based on the tone mapping method of quick bilateral filtering, has proposed a kind of tone mapping method that keeps full Variation Model based on the edge.

The invention provides and a kind ofly keep the tone mapping method of full sub-model based on the edge, comprising:

Input a panel height dynamic image;

High-dynamics image to input carries out light intensity reconstruct;

Obtain the logarithm value of the light intensity value of neotectonics;

Utilize the edge to keep full variation wave filter that the light intensity logarithm value that obtains is carried out filtering and obtain the image basis layer;

Utilize light intensity logarithm value subtracted image basal layer to obtain the levels of detail of image;

Calculate the compression factor factor and image output light intensity;

Obtain a new scale factor M with light intensity after the compression divided by the input picture light intensity.Scale factor M is acted on respectively on red channel, green channel, each passage of blue channel, obtain compressing the image behind the color;

Carry out gamma and proofread and correct and obtain low dynamic image obtaining compressing image behind the color.

Described high-dynamics image to input carries out light intensity reconstruct and comprises:

According to the channel value of red channel R, green channel G, blue channel B, utilize formula intensity=0.299*R+0.587*G+0.114*B, re-construct out light intensity intensity value.

The described calculating compression factor factor and image output light intensity comprise:

Calculate the compression factor factor: utilize formula compressfactor=log (outputrange)/(max (log_base)-min (log_base)) * 1.5, wherein outputrange is the dynamic range of images of output, max (log_base) is the maximal value of getting all pixel values among the image basis layer log_base, min (log_base) is the minimum value of getting all pixel values among the image basis layer log_base, and 1.5 is empirical value;

Computed image output light intensity: utilize formula output_intensity=exp (compressfactor*log_base+log_detail).

Described with compressing rear light intensity divided by new scale factor M of input picture light intensity acquisition.Scale factor M is acted on respectively on red channel, green channel, each passage of blue channel, and the image that obtains compressing behind the color comprises:

Utilize formula

$M=\frac{\mathrm{output}\_\mathrm{intensity}}{\mathrm{intensity}},$

R, G, three color channel compression factor of B factor have been obtained; Value after R, G, the B compression is easy to be calculated;

$\left\{\begin{array}{c}R\_\mathrm{Output}=M*R+\mathrm{Input}\\ G\_\mathrm{Output}=M*G\_\mathrm{Input}\\ B\_\mathrm{Output}=M*B\_\mathrm{Input}\end{array}\right.,$

Wherein R_Output, G_Output, B_Output are respectively compressed images R, G, B color-values, R_Input, G_Input, B_Input are respectively input picture R, G, B color-values, the calculating of doing all is on the logarithm of image light intensity, the difference of this each pixel value of sampled images obtains unified the processing to whole image range simultaneously with regard to the contrast of correspondence image.Above technology can be found out, keeps the tone mapping method of full Variation Model can obtain the very high low dynamic image of quality based on the edge, and the treatment effect of color and contrast is very good, and can not produce halation phenomenon.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art, the below will do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art, apparently, accompanying drawing in the following describes only is some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain according to these accompanying drawings other accompanying drawing.

Fig. 1 keeps the tone mapping method process flow diagram of full Variation Model based on the edge in the embodiment of the invention.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the invention, the technical scheme in the embodiment of the invention is clearly and completely described, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills belong to the scope of protection of the invention not making all other embodiment that obtain under the creative work prerequisite.

The present invention has overcome the shortcoming that can produce halation phenomenon based on the tone mapping method of quick bilateral filtering, has proposed a kind of tone mapping method that keeps full Variation Model based on the edge.Total variation (Total Variation) regularization model 1992 by Rudin, Osher and Fatemi propose, with the noise that removes in the image.This model has two base attributes: one is the edge that can keep image, and treatment effect is very good under given conditions; Another is that variable quantity and the image detail yardstick of image light intensity is inversely proportional to.More succinct says, this model keep the edge information very effectively in smooth grain.This can realize by minimizing equation.

$u=\underset{u}{\arg \min }\{\frac{1}{2}{\left|\right|f-\mathrm{Ku}\left|\right|}^2+\mathrm{\λ}{\left|u\right|}_{\mathrm{TV}}\}---\left(1\right)$

Wherein first is the fidelity item, and the image u after assurance is level and smooth keeps the principal character of observing image f; Second is regular terms, reaches the purpose of smoothed image by the total variation that minimizes u.λ＞0th, scale factor plays the effect of balance fidelity item and regular terms.The value of λ is larger, and smooth effect is more obvious.K is the linear operator that determinacy is degenerated.The Euler-Lagrange equation that formula (1) is corresponding is

$K^{*}(f-\mathrm{Ku})+\mathrm{\λ}\mathrm{div}\left(\frac{\▿u}{|\▿u|}\right)=0,---\left(2\right)$

K wherein ^*It is the adjoint operator of K.Formula (1) has identical solution with formula (2), and can be in the hope of unique solution.

Adopt method of steepest descent to find the solution to formula (1), add suitable first, boundary value condition, then be constructed as follows reaction-diffusion equation,

$\left\{\begin{array}{c}\frac{\∂u}{\∂t}(t;x)=K*(f\left(x\right)-\mathrm{Ku}(t;x))+\mathrm{\λ}\mathrm{div}\left(\frac{\▿u}{|\▿u|}\right)(t;x),\\ \begin{array}{cc}& (t;x)\∈(0,T]\×\mathrm{\Ω}\\ u(0,x)=f\left(x\right)& x\∈\mathrm{\Ω}\\ \frac{\∂u}{\∂n}(t,x){|}_{\∂\mathrm{\Ω}}=0& (t;x)\∈(0,T]\×\mathrm{\Ω}\end{array}\end{array}\right.,---\left(3\right)$

Wherein,

Be the border of Ω, n is the border

On unit outside normal vector.Formula (3) is adopted finite difference method.

Utilize edge recited above to keep operator, we are easy to realize two Scale Decompositions of image.Resolve into the levels of detail of the basal layer of sectionally smooth, high-contrast and low contrast, small scale texture.Basal layer keeps the total variation wave filter to obtain by the edge, deducts basal layer with input picture and just obtains levels of detail.In particular, g represents a width of cloth input picture and is used to two Scale Decompositions, f _TVFor the edge keeps function of total variation, b is basal layer, and d is levels of detail, and said process is expressed as

b＝f _TV(g), (4)

d=g-b. (5)

The purpose of tone mapping is that a panel height dynamic image is converted into the low dynamic image that a width of cloth can show in conventional display apparatus.In this process, must reduce contrast, reach the dynamic range that conventional display apparatus can show.Simultaneously, in order to obtain a high-quality low dynamic image, the details in the necessary former high-dynamics image of reservation as much as possible.Basal layer comprises high-contrast information, and levels of detail comprises detail textures information.So only need basal layer is carried out the yardstick compression, levels of detail remains unchanged, and just can reach to reduce the purpose that contrast keeps details simultaneously.At last, basal layer and levels of detail synthesize new images after will compressing, and are needed LDR image.It is as follows to be expressed as the formula form

u=C*b+d, (6)

Wherein, C is the compression factor factor, C＜1, and u is the low dynamic image after synthetic.Bringing formula (5) into formula (6) gets

u＝g-(1-C)*b, (7)

By formula (7), see that more significantly only the large scale of image is compressed, other small scale details are retained.Can regulate the luminance brightness compression degree by changing scale factor C, to satisfy the human visual effect of being wanted.Bring formula (4) into formula (7) and just obtained the tone mapping method that this patent proposes

u＝g-(1-C)*f _TV(g). (8)

For the processing of color of image, we at first calculate the light intensity after the compression, have obtained so a new scale factor M with light intensity after the compression divided by the input picture light intensity.Scale factor M is acted on respectively on R, G, the B passage, just obtain compressing the image behind the color.In particular, I_Input is the input picture light intensity, and I_Output is light intensity after the compression of images.Then

$M=\frac{I\_\mathrm{Output}}{I\_\mathrm{Input}},---\left(9\right)$

R, G, three color channel compression factor of B factor have been obtained like this.Value after R, G, the B compression is easy to be calculated.As follows

$\left\{\begin{array}{c}R\_\mathrm{Output}=M*R+\mathrm{Input}\\ G\_\mathrm{Output}=M*G\_\mathrm{Input}\\ B\_\mathrm{Output}=M*B\_\mathrm{Input}\end{array}\right.,---\left(10\right)$

Wherein R_Output, G_Output, B_Output are respectively compressed images R, G, B color-values, and R_Input, G_Input, B_Input are respectively input picture R, G, B color-values.The calculating that we do all is on the logarithm of image light intensity, and the difference of this each pixel value of sampled images can obtain unified the processing to whole image range simultaneously with regard to the contrast of correspondence image.

At last image is carried out gamma correction.The method can obtain the very high low dynamic image of quality, and the treatment effect of color and contrast is very good, and can not produce halation phenomenon.

Illustrate and keep the tone mapping method of full sub-model based on the edge in the embodiment of the invention, concrete steps are as follows:

S101: input a panel height dynamic image.

S102: carry out light intensity (intensity) reconstruct.According to the channel value of red channel (R), green channel (G), blue channel (B), utilize formula intensity=0.299*R+0.587*G+0.114*B, re-construct out light intensity value.

S103: the logarithm value (log_intensity) that obtains the light intensity value of neotectonics.

S104: image filtering.The light intensity logarithm value log_intensity that utilizes the edge to keep full variation wave filter that step 3 is obtained carries out filtering and obtains image basis layer (log_base).

S105: picture breakdown.Utilize the light intensity logarithm value log_intensity of step S103 to deduct the image basis layer log_base of step S104, obtain the levels of detail (log_detail) of image.So just a panel height dynamic image is decomposed into image basis layer and image detail layer.

S106: calculate compression factor factor compressfactor.Utilize formula compressfactor=log (outputrange)/(max (log_base)-min (log_base)) * 1.5, wherein outputrange is the dynamic range of images of output, max (log_base) is the maximal value of getting all pixel values among the image basis layer log_base, min (log_base) is the minimum value of getting all pixel values among the image basis layer log_base, and 1.5 is empirical value.

S107: computed image output light intensity output_intensity.Utilize formula output_intensity=exp (compressfactor*log_base+log_detail).

S108: color treatments.Calculate light intensity output_intensity after the compression by step S107, obtained so a new scale factor M with light intensity after the compression divided by input picture light intensity intensity.Scale factor M is acted on respectively on red channel (R), green channel (G), each passage of blue channel (B), just obtain compressing the image behind the color.See that formula is as follows:

$M=\frac{\mathrm{output}\_\mathrm{intensity}}{\mathrm{intensity}},$

R, G, three color channel compression factor of B factor have been obtained like this.Value after R, G, the B compression is easy to be calculated.As follows

$\left\{\begin{array}{c}R\_\mathrm{Output}=M*R+\mathrm{Input}\\ G\_\mathrm{Output}=M*G\_\mathrm{Input}\\ B\_\mathrm{Output}=M*B\_\mathrm{Input}\end{array}\right.,$

Wherein R_Output, G_Output, B_Output are respectively compressed images R, G, B color-values, and R_Input, G_Input, B_Input are respectively input picture R, G, B color-values.The calculating that we do all is on the logarithm of image light intensity, and the difference of this each pixel value of sampled images can obtain unified the processing to whole image range simultaneously with regard to the contrast of correspondence image.

S109:gamma proofreaies and correct.So just obtained low dynamic image.

To sum up, keep the tone mapping method of full Variation Model can obtain the very high low dynamic image of quality based on the edge, the treatment effect of color and contrast is very good, and can not produce halation phenomenon.

One of ordinary skill in the art will appreciate that all or part of step in the whole bag of tricks of above-described embodiment is to come the relevant hardware of instruction finish by program, this program can be stored in the computer-readable recording medium, storage medium can comprise: ROM (read-only memory) (ROM, Read Only Memory), random access memory (RAM, Random AccessMemory), disk or CD etc.

More than keep the tone mapping method of full Variation Model to what the embodiment of the invention provided based on the edge, be described in detail, used specific case herein principle of the present invention and embodiment are set forth, the explanation of above embodiment just is used for helping to understand method of the present invention and core concept thereof; Simultaneously, for one of ordinary skill in the art, according to thought of the present invention, all will change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.

Claims (4)

1. one kind keeps the tone mapping method of full sub-model based on the edge, it is characterized in that, comprising:

Input a panel height dynamic image;

High-dynamics image to input carries out light intensity reconstruct;

Obtain the logarithm value of the light intensity value of neotectonics;

Utilize the edge to keep full variation wave filter that the light intensity logarithm value that obtains is carried out filtering and obtain the image basis layer;

Utilize light intensity logarithm value subtracted image basal layer to obtain the levels of detail of image;

Calculate the compression factor factor and image output light intensity;

Obtain a new scale factor M with light intensity after the compression divided by the input picture light intensity.Scale factor M is acted on respectively on red channel, green channel, each passage of blue channel, obtain compressing the image behind the color;

Carry out gamma and proofread and correct and obtain low dynamic image obtaining compressing image behind the color.

2. as claimed in claim 1ly keep the tone mapping method of full sub-model based on the edge, it is characterized in that described high-dynamics image to input carries out light intensity reconstruct and comprises:

According to the channel value of red channel R, green channel G, blue channel B, utilize formula intensity=0.299*R+0.587*G+0.114*B, re-construct out light intensity intensity value.

3. as claimed in claim 2ly keep the tone mapping method of full sub-model based on the edge, it is characterized in that the described calculating compression factor factor and image output light intensity comprise:

Calculate the compression factor factor: utilize formula compressfactor=log (outputrange)/(max (log_base)-min (log_base)) * 1.5, wherein outputrange is the dynamic range of images of output, max (log_base) is the maximal value of getting all pixel values among the image basis layer log_base, min (log_base) is the minimum value of getting all pixel values among the image basis layer log_base, and 1.5 is empirical value;

Computed image output light intensity: utilize formula output_intensity=exp (compressfactor*log_base+log_detail).

4. as claimed in claim 3ly keep the tone mapping method of full sub-model based on the edge, it is characterized in that, describedly obtain a new scale factor M with light intensity after the compression divided by the input picture light intensity.Scale factor M is acted on respectively on red channel, green channel, each passage of blue channel, and the image that obtains compressing behind the color comprises:

Utilize formula

$M=\frac{\mathrm{output}\_\mathrm{intensity}}{\mathrm{intensity}},$

R, G, three color channel compression factor of B factor have been obtained; Value after R, G, the B compression is easy to be calculated;

$\left\{\begin{array}{c}R\_\mathrm{Output}=M*R+\mathrm{Input}\\ G\_\mathrm{Output}=M*G\_\mathrm{Input}\\ B\_\mathrm{Output}=M*B\_\mathrm{Input}\end{array}\right.,$

Wherein R_Output, G_Output, B_Output are respectively compressed images R, G, B color-values, R_Input, G_Input, B_Input are respectively input picture R, G, B color-values, the calculating of doing all is on the logarithm of image light intensity, the difference of this each pixel value of sampled images obtains unified the processing to whole image range simultaneously with regard to the contrast of correspondence image.

Images