CN104463794A - Image processing method based on partitions - Google Patents

Abstract

The invention provides an image processing method based on partitions. The image processing method comprises the steps that an original floating-point type image is transformed into an XYZ space from an RGB space; the image in the XYZ space is partitioned according to the brightness of the image; non-local average filtering is conducted on the brightness of the image, and a basic layer and a detail layer of the image in the XYZ space are obtained; color adaptation transformation and tone compression are conducted on the base layer of the image in the XYZ space; an RGB image of the base layer is transformed into the XYZ space; the XYZ image of the base layer and the detail layer are combined and then transformed into a uniform color space; attribute regulation is conducted on detail attributes and image attributes of the image in the uniform color space; the image in the uniform color space is transformed into the RGB space and then output. The image processing method has the advantages that details can be well reserved and halos are not generated, and the image processing method can be widely applied to the field of image processing.

Description

A kind of image processing method based on subregion

Technical field

The present invention relates to image processing techniques, particularly relate to a kind of image processing method based on subregion.

Background technology

In recent years, high dynamic range images (HDRI, high dynamic range image) receive the concern of increasing people, and be widely used in the fields such as digital photography, art of film, scientific imagery enhancing, virtual reality, mutual 3D application.Here, the luminance brightness contrast of the dynamic range of images brightest area that refers to image and most dark areas.There is dynamic range widely in occurring in nature, e.g., spans the dynamic range of 8 orders of magnitude from faint starlight to sunshine.In practical application, conventional display apparatus can only show the low-dynamic range scene of 2 orders of magnitude.

In order to project on display device by truly complete for high dynamic range images, researchist proposes tone mapping method, high dynamic range data compression is mapped in the displayable contrast range of conventional display apparatus by it, reduces the loss of image information in details, color, contrast and lightness etc. simultaneously.Tone mapping method is divided into global map, local maps, colored quantum noise three class.The overall contrast of global map to image significantly decays, and calculates fairly simple, but easily loses the information of detail section.The local relation be mapping through between pixel and pixel suitably strengthens the luminance contrast of subrange, and it remains certain details, but some region there will be the phenomenon of distortion, halation.The effect that colored quantum noise method carries out tone mapping to high dynamic range images is better, but owing to have employed gaussian filtering, bilateral filtering in model, therefore easily produce obvious halation phenomenon.

As can be seen here, in the prior art, there is the easily problem such as loss details or generation halation in the image processing method based on tone mapping method.

Summary of the invention

In view of this, fundamental purpose of the present invention is to provide a kind of and can retains details preferably and not produce the image processing method based on subregion of halation.

In order to achieve the above object, the technical scheme that the present invention proposes is:

Based on an image processing method for subregion, comprise the steps:

Step 1, original floating-point type image is transformed into XYZ space according to following conversion regime by rgb space, obtains XYZ space image:

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=M_{\mathrm{sRGB}}\left[\begin{array}{c}R\\ G\\ B\end{array}\right],M_{\mathrm{sRGB}}=\left[\begin{array}{ccc}0.4124& 0.2127& 0.0193\\ 0.3576& 0.7152& 0.1192\\ 0.1805& 0.0722& 0.9504\end{array}\right];$

Wherein, X, Z all represent chromatic component, and Y=(D (ζ) | ζ ∈ Y} represents brightness of image, and ζ represents location of pixels; M _sRGBrepresent color space conversion matrix.

Step 2, the low photo threshold of brightness of image, brightness of image height photo threshold are set according to brightness of image average, and according to brightness of image, subregion is carried out to XYZ space image, obtain shadow region, mesozone, specular.

Step 3, subregion based on step 2, to brightness of image Y={D (ζ) | ζ ∈ Y} carries out non-local mean filtering, obtains the Primary layer RNL [D (ζ)] of XYZ space image.

The levels of detail of step 4, acquisition XYZ space image: d=Y-RNL [D (ζ)].

Step 5, successively chromatic adaptation conversion is carried out to the Primary layer of XYZ space image and tone compresses, obtain Primary layer RGB image.

Step 6, Primary layer RGB image is transformed into XYZ space, obtains Primary layer XYZ image [X _cy _cz _c] ^t; And be transformed into uniform colour space after Primary layer XYZ image and levels of detail being merged, obtain uniform colour space image.

Step 7, attribute adjustment is carried out to the detail attribute of uniform colour space image and image attributes.

Step 8, by color adaption inverse transformation, XYZ space is converted into uniform colour space image after, then be converted to rgb space, output image.

In sum, after first original floating-point type image is transformed into XYZ space by rgb space by the image processing method based on subregion of the present invention, brightness of image average according to the XYZ space image obtained carries out subregion, and according to subregion, non-local mean filtering is carried out to brightness of image, thus XYZ space image is divided into Primary layer and levels of detail.Secondly, the Primary layer RGB image carrying out the Primary layer of XYZ space image obtaining after chromatic adaptation conversion and tone compress is transformed into XYZ space by the image processing method based on subregion of the present invention, and is transformed into uniform colour space after Primary layer XYZ image and levels of detail being merged; Output image after rgb space is converted to uniform colour space image.The inventive method only carries out filtering, chromatic adaptation process, tone process to XYZ space image Primary layer, but not this type of process is carried out to whole XYZ space image, therefore the inventive method can retain the minutia of image more, so just maintain image boundary better, avoid the generation of halation phenomenon.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the image processing method based on subregion of the present invention.

Fig. 2 is the schematic flow sheet carrying out subregion according to brightness of image of the present invention.

Fig. 3 is the schematic flow sheet being transformed into uniform colour space after Primary layer XYZ image of the present invention and levels of detail merge.

Fig. 4 (a) is original floating-point type image.

Fig. 4 (b) is the image obtained after adopting the image processing method based on subregion of the present invention to process original floating-point type image.

Embodiment

For making the object, technical solutions and advantages of the present invention clearly, the present invention is described in further detail below in conjunction with the accompanying drawings and the specific embodiments.

Fig. 1 is the process flow diagram of the image processing method based on subregion of the present invention.As shown in Figure 1, the image processing method based on subregion of the present invention, comprises the steps:

Step 1, original floating-point type image is transformed into XYZ space according to following conversion regime by rgb space, obtains XYZ space image:

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=M_{\mathrm{sRGB}}\left[\begin{array}{c}R\\ G\\ B\end{array}\right],M_{\mathrm{sRGB}}=\left[\begin{array}{ccc}0.4124& 0.2127& 0.0193\\ 0.3576& 0.7152& 0.1192\\ 0.1805& 0.0722& 0.9504\end{array}\right];$

Wherein, X, Z all represent chromatic component, and Y=(D (ζ) | ζ ∈ Y} represents brightness of image, and ζ represents location of pixels; M _sRGBrepresent color space conversion matrix.

In the inventive method, color space conversion matrix M _sRGBbe retrieved as prior art, repeat no more herein.In practical application, the image adopting rgb space to represent is easily by the impact of display device performance, and the image adopting XYZ space to represent is not by the impact of display device performance.

Step 2, the low photo threshold of brightness of image, brightness of image height photo threshold are set according to brightness of image average, and according to brightness of image, subregion is carried out to XYZ space image, obtain shadow region, mesozone, specular.

Step 3, subregion based on step 2, to brightness of image Y={D (ζ) | ζ ∈ Y} carries out non-local mean filtering, obtains the Primary layer RNL [D (ζ)] of XYZ space image.

The levels of detail of step 4, acquisition XYZ space image: d=Y-RNL [D (ζ)].

Step 5, successively chromatic adaptation conversion is carried out to the Primary layer of XYZ space image and tone compresses, obtain Primary layer RGB image.

In the present invention, the compression of chromatic adaptation transform domain tone is prior art, repeats no more herein.

Step 6, Primary layer RGB image is transformed into XYZ space, obtains Primary layer XYZ image [X _cy _cz _c] ^t; And be transformed into uniform colour space after Primary layer XYZ image and levels of detail being merged, obtain uniform colour space image.

Here, uniform colour space is IPT space; Wherein, I represents lightness axis, and P represents red-green axle, and T represents Huang-blue axle.

Step 7, attribute adjustment is carried out to the detail attribute of uniform colour space image and image attributes.

Step 8, by color adaption inverse transformation, XYZ space is converted into uniform colour space image after, then be converted to rgb space, output image.

In the inventive method, XYZ space is International Commission on Illumination (CIE, International Commissionon Illumination) XYZ space.

In a word, after first original floating-point type image is transformed into XYZ space by rgb space by the image processing method based on subregion of the present invention, brightness of image average according to the XYZ space image obtained carries out subregion, and according to subregion, non-local mean filtering is carried out to brightness of image, thus XYZ space image is divided into Primary layer and levels of detail.Secondly, the Primary layer RGB image carrying out the Primary layer of XYZ space image obtaining after chromatic adaptation conversion and tone compress is transformed into XYZ space by the image processing method based on subregion of the present invention, and is transformed into uniform colour space after Primary layer XYZ image and levels of detail being merged; Output image after rgb space is converted to uniform colour space image.The inventive method only carries out filtering, chromatic adaptation process, tone process to XYZ space image Primary layer, but not this type of process is carried out to whole XYZ space image, therefore the inventive method can retain the minutia of image more, so just maintain image boundary better, avoid the generation of halation phenomenon.

Fig. 2 is the schematic flow sheet carrying out subregion according to brightness of image of the present invention.As shown in Figure 2, in the inventive method, described step 2 is specially:

Step 21, acquisition brightness of image average $k=\frac{Y_{\mathrm{av}}-\lg Y_{\min }}{\lg Y_{\max }-\lg Y_{\min }};$ Wherein, $Y_{\mathrm{av}}=(1/W)\underset{\mathrm{\ζ}\∈Y}{\mathrm{\Σ}}\lg [\mathrm{\δ}+D\left(\mathrm{\ζ}\right)];$ δ is parameter, and 10 ^-6≤ δ≤10 ^-2; Y _maxrepresent high-high brightness; Y _minrepresent minimum brightness; W represents the brightness of image sum of all pixels of XYZ space image.

Step 22, the low photo threshold L of brightness of image is set _t1=L _max-(0.9+0.1k) (L _max-L _min), brightness of image height photo threshold L _t2=L _min+ [0.6+0.4 (01-k)] (L _max-L _min); Wherein, L _max, L _minbe respectively the maximal value after brightness of image Y normalization, minimum value, and

Step 23, brightness of image average k < L _t1xYZ space image-region be shadow region, brightness of image average L _t1< k < L _t2xYZ space image-region be mesozone, brightness of image average k > L _t2xYZ space image-region be specular.

In practical application, after carrying out subregion according to brightness of image, pixel XYZ space image belonging to same subregion can be comparatively near apart, also can be apart from each other; That is, other pixels between two pixels belonging to same subregion can be the pixels belonging to another subregion.

In the inventive method, in step 3, described to brightness of image Y={D (ζ) | ζ ∈ Y} carries out non-local mean filtering, obtains the Primary layer RNL [D (ζ)] of XYZ space image, is specially:

$\mathrm{RNL}\left[D\left(\mathrm{\&zeta;}\right)\right]=\underset{\mathrm{\&theta;}\&Element;Y}{\mathrm{\&Sigma;}}{w}_{\mathrm{RNL}}(\mathrm{\&zeta;},\mathrm{\&theta;})D\left(\mathrm{\&theta;}\right);$

Wherein, θ represents the location of pixels being different from ζ, subregion coefficient weight $w_s(\mathrm{\&zeta;},\mathrm{\&theta;})=e^{\frac{{\left|\right|G\left[D\left(\mathrm{\&zeta;}\right)\right]-G\left[D\left(\mathrm{\&theta;}\right)\right]\left|\right|}^2}{h^2}};$ H represents XYZ space image smoothing parameter; The rectangle field vector that the rectangle field vector that G [D (ζ)] is ζ, G [D (θ)] are θ.

In the inventive method, the Primary layer RNL{D (ζ) of XYZ space image] be compared to XYZ space figure, its change scale ratio is greatly.

Fig. 3 is the schematic flow sheet being transformed into uniform colour space after Primary layer XYZ image of the present invention and levels of detail merge.As shown in Figure 3, in step 6, described Primary layer XYZ image and levels of detail are merged after be transformed into uniform colour space, obtain uniform colour space image, be specially:

Step 61, just Primary layer XYZ image and levels of detail merge, and obtain merging image:

$\left[\begin{array}{c}A\\ V\\ S\end{array}\right]=M_H^{D65}\left[\begin{array}{c}{X}_c\\ Y_c\\ Z_c\end{array}\right],M_H^{D65}=\left[\begin{array}{ccc}0.4002& 0.7075& -0.0807\\ -0.2280& 1.1500& 0.0612\\ 0.0000& 0.0000& 0.9184\end{array}\right].$

Step 62, be combined image and carry out index replacement: A '=A ^0.43, V '=V ^0.43, S '=S ^0.43.

Step 63, the merging image through index replacement is transformed into uniform colour space, obtains uniform colour space image:

$\left[\begin{array}{c}I\\ P\\ T\end{array}\right]=M_{\mathrm{IPT}}\left[\begin{array}{c}{A}^{\&prime;}\\ V^{\&prime;}\\ S^{\&prime;}\end{array}\right],M_{\mathrm{IPT}}=\left[\begin{array}{ccc}0.4000& 0.4000& 0.2000\\ 4.4550& -4.8510& 0.3960\\ 0.8056& 0.3572& -1.1628\end{array}\right].$

In the inventive method, in step 7, the attribute of described detail attribute is adjusted to: wherein, F _lfor auto-adaptive parameter.

In the inventive method, in step 7, the attribute of described image attributes is adjusted to:

$P=P\&CenterDot;[{(F_L+1)}^{0.2}\&CenterDot;\frac{1.29C^2-0.27C+0.42}{C^2-0.31C+0.42}]Y-\mathrm{RNL}\left[D\left(\mathrm{\&zeta;}\right)\right]$

$T=T\&CenterDot;[{(F_L+1)}^{0.2}\&CenterDot;\frac{1.29C^2-0.27C+0.42}{C^2-0.31C+0.42}]Y-\mathrm{RNL}\left[D\left(\mathrm{\&zeta;}\right)\right]$

I _a＝I ^γ；

Wherein, C is color variables; γ is adjustment parameter, and γ=1.5 when uniform colour space image pixel value is 0; Uniform colour space image pixel value be (0,20] time γ=1.25; Uniform colour space image pixel value be (20,100] time γ=1.0.

In practical application, represent that when uniform colour space image pixel value is 0 uniform colour space color of image is black, uniform colour space image pixel value is (0,20] represent time that uniform colour space color of image is grey, uniform colour space image pixel value be (20,100] time represent that uniform colour space color of image is average color.

Fig. 4 (a) is original floating-point type image.Fig. 4 (b) is the image obtained after adopting the image processing method based on subregion of the present invention to process original floating-point type image.As shown in Fig. 4 (a), Fig. 4 (b), the image processing method based on subregion of the present invention remains the detailed information of original floating-point type image substantially; Meanwhile, the image boundary obtained after process is clear, and can not produce halation phenomenon, and effect is better.

In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (5)

1. based on an image processing method for subregion, it is characterized in that, described image processing method comprises the steps:

Step 1, original floating-point type image is transformed into XYZ space according to following conversion regime by rgb space, obtains XYZ space image:

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=M_{\mathrm{sRGB}}\left[\begin{array}{c}R\\ G\\ B\end{array}\right],M_{\mathrm{sRGB}}=\left[\begin{array}{ccc}0.4124& 0.2127& 0.0193\\ 0.3576& 0.7152& 0.1192\\ 0.1805& 0.0722& 0.9504\end{array}\right];$

Wherein, X, Z all represent chromatic component, Y={D (ζ) | ζ ∈ Y} represents brightness of image, and ζ represents location of pixels; M _sRGBrepresent color space conversion matrix;

Step 2, the low photo threshold of brightness of image, brightness of image height photo threshold are set according to brightness of image average, and according to brightness of image, subregion is carried out to XYZ space image, obtain shadow region, mesozone, specular;

Step 3, subregion based on step 2, to brightness of image Y={D) ζ) | ζ ∈ Y} carries out non-local mean filtering, obtains the Primary layer RNL [D (ζ)] of XYZ space image;

The levels of detail of step 4, acquisition XYZ space image: d=Y-RNL [D (ζ)];

Step 5, successively chromatic adaptation conversion is carried out to the Primary layer of XYZ space image and tone compresses, obtain Primary layer RGB image;

Step 6, Primary layer RGB image is transformed into XYZ space, obtains Primary layer XYZ image [X _cy _cz _c] ^t; And be transformed into uniform colour space after Primary layer XYZ image and levels of detail being merged, obtain uniform colour space image;

Step 7, attribute adjustment is carried out to the detail attribute of uniform colour space image and image attributes;

Step 8, by color adaption inverse transformation, XYZ space is converted into uniform colour space image after, then be converted to rgb space, output image.

2. the image processing method based on subregion according to claim 1, is characterized in that, described step 2 is specially:

Step 21, acquisition brightness of image average $k=\frac{Y_{\mathrm{av}}-\lg Y_{\min }}{\lg Y_{\max }-\lg Y_{\min }};$ Wherein, $Y_{\mathrm{av}}=(1/W)\underset{\mathrm{\&zeta;}\&Element;Y}{\mathrm{\&Sigma;}}\lg [\mathrm{\&delta;}+D\left(\mathrm{\&zeta;}\right)];$ δ is parameter, and 10 ^-6≤ δ≤10 ^-2; Y _maxrepresent high-high brightness; Y _minrepresent minimum brightness; W represents the brightness of image sum of all pixels of XYZ space image;

Step 22, the low photo threshold L of brightness of image is set _t1=L _max-(0.9+0.1k) (L _max-L _min), brightness of image height photo threshold L _t2=L _min+ [0.6+0.4 (01-k)] (L _max-L _min); Wherein, L _max, L _minbe respectively the maximal value after brightness of image Y normalization, minimum value, and L _max=1,

Step 23, brightness of image average k < L _t1xYZ space image-region be shadow region, brightness of image average L _t1< k < L _t2xYZ space image-region be mesozone, brightness of image average k < L _t2xYZ space image-region be specular.

3. the image processing method based on subregion according to claim 1, it is characterized in that, in step 3, described to brightness of image Y=(D (ζ) | ζ ∈ Y} carries out non-local mean filtering, obtain the Primary layer RNL [D (ζ)] of XYZ space image, be specially:

$\mathrm{RNL}\left[D\left(\mathrm{\&zeta;}\right)\right]=\underset{\mathrm{\&theta;}\&Element;Y}{\mathrm{\&Sigma;}}{w}_{\mathrm{RNL}}(\mathrm{\&zeta;},\mathrm{\&theta;})D\left(\mathrm{\&theta;}\right);$

Wherein, represent the location of pixels being different from ζ, subregion coefficient weight $w_s(\mathrm{\&zeta;},\mathrm{\&theta;})=e^{\frac{{\left|\right|G\left[D\left(\mathrm{\&zeta;}\right)\right]-G\left[D\left(\mathrm{\&theta;}\right)\right]\left|\right|}^2}{h^2}};$ H represents XYZ space image smoothing parameter; The rectangle field vector that G [D (ζ)] is ζ, for rectangle field vector.

4. the image processing method based on subregion according to claim 1, is characterized in that, in step 6, described Primary layer XYZ image and levels of detail are merged after be transformed into uniform colour space, obtain uniform colour space image, be specially:

Step 61, just Primary layer XYZ image and levels of detail merge, and obtain merging image:

$\left[\begin{array}{c}A\\ V\\ S\end{array}\right]=M_H^{D65}\left[\begin{array}{c}{X}_c\\ Y_c\\ Z_c\end{array}\right],M_H^{D65}=\left[\begin{array}{ccc}0.4002& 0.7075& -0.0807\\ -0.2280& 1.1500& 0.0612\\ 0.0000& 0.0000& 0.9184\end{array}\right];$

Step 62, be combined image and carry out index replacement: A '=A ^0.43, V '=V ^0.43, S '=S ^0.43;

Step 63, the merging image through index replacement is transformed into uniform colour space, obtains uniform colour space image:

$\left[\begin{array}{c}I\\ P\\ T\end{array}\right]=M_{\mathrm{IPT}}\left[\begin{array}{c}{A}^{\&prime;}\\ V^{\&prime;}\\ S^{\&prime;}\end{array}\right],M_{\mathrm{IPT}}=\left[\begin{array}{ccc}0.4000& 0.4000& 0.2000\\ 4.4550& -4.8510& 0.3960\\ 0.8056& 0.3572& -1.1628\end{array}\right].$

5. the image processing method based on subregion according to claim 1, is characterized in that, in step 7, the attribute of described detail attribute is adjusted to: wherein, F _lfor auto-adaptive parameter;

The attribute of described image attributes is adjusted to:

$P=P\&CenterDot;[{(F_L+1)}^{0.2}\&CenterDot;\frac{1.29C^2-0.27C+0.42}{C^2-0.31C+0.42}]Y-\mathrm{RNL}\left[D\left(\mathrm{\&zeta;}\right)\right]$

$T=T\&CenterDot;[{(F_L+1)}^{0.2}\&CenterDot;\frac{1.29C^2-0.27C+0.42}{C^2-0.31C+0.42}]Y-\mathrm{RNL}\left[D\left(\mathrm{\&zeta;}\right)\right]$

I _a＝I ^γ；

Wherein, C is color variables; γ is adjustment parameter, and γ=1.5 when uniform colour space image pixel value is 0; Uniform colour space image pixel value be (0,20] time γ=1.25; Uniform colour space image pixel value be (20,100] time γ=1.0.