CN110415188A - A tone mapping method for HDR images based on multi-scale morphology - Google Patents

Abstract

The invention discloses a multi-scale morphology-based HDR image tone mapping method, the method comprising: S1, inputting the HDR image to be compressed; S2, constructing the logarithmic image of the HDR image to be compressed; S3, performing the logarithmic image processing on the logarithmic image Dynamic range compression and multi-scale decomposition to obtain the first Gaussian pyramid and the first Laplacian pyramid; S4, using a one-dimensional boundary-preserving polishing operator to extract the enhanced detail layer of the first Gaussian pyramid, and the details of the first Gaussian pyramid Add layer by layer to the layer corresponding to the first Laplacian pyramid to obtain the second Laplacian pyramid; S5, perform dynamic range compression reconstruction on the second Laplacian pyramid, and obtain detail enhancement after dynamic range compression image; S6, performing exponential transformation on the detail-enhanced image to obtain a first LDR image; S7, performing color correction on the first LDR image to obtain an LDR image to be output. The invention can realize effective compression of the dynamic range of the HDR image, and can meet certain real-time application requirements.

Description

A tone mapping method for HDR images based on multi-scale morphology

technical field

The present invention relates to the technical field of image processing, in particular to a multi-scale morphology-based HDR image tone mapping method.

Background technique

Compared with LDR (English full name is "Low Dynamic Range", Chinese full name is "low dynamic range") images, HDR (English full name is "High Dynamic Range", Chinese full name is "High Dynamic Range") images can express more scene details Contrast information with brightness has been widely used in digital image technology. Dedicated HDR display devices are expensive and cost-effective. When an HDR image is displayed on an LDR display device, it cannot truly reproduce the complete light and shadow effects of the original scene, but a good tone mapping method can achieve effective compression of the dynamic range of the HDR image and solve the problem of dynamic range mismatch to meet more practical applications need.

Generally speaking, methods related to tone mapping in the field of HDR images are mainly divided into two categories: global tone mapping methods and local tone mapping methods. Among them, the global tone mapping method uses the same mapping function to transform each pixel in the HDR image. This type of method applies the same mapping function to each pixel, which is a one-to-one mapping relationship, so the advantages of this type of method are that the method is simple and the operation speed is fast, but this method does not consider the spatial position of the pixel point, only Considers the grayscale value of the pixel, resulting in a loss of brightness, color, and detail in the resulting image. As early as 1984, Miller proposed a global method based on Stevens psychophysical experimental data. In 1993, Tumblin and Rushmeier also proposed a global method for the brightness domain based on Stevens' psychophysical experiments, which is nonlinear. In 1994, based on visual sensitivity, Ward proposed a linear mapping method that can improve the contrast and brightness of images. In 2003, Drago proposed a tone mapping method based on the logarithmic transformation, because the logarithmic transformation is more similar to the human eye's perception of light, but the method loses serious details and cannot efficiently compress the dynamic range.

The local tone mapping method uses different mapping functions to transform different pixels in the HDR image. Different transformations are performed on different areas where each pixel is located, and the spatial position of the pixel is also taken into consideration. In the end, it is possible that different pixel values before mapping become the same value after mapping, while the pixel values before mapping are the same The situation where the points become different values due to different mappings of spatial positions. Compared with the global tone mapping method, the local tone mapping method not only considers the gray value of the pixel, but also further considers the spatial position of the pixel, so the resulting image processed by this type of method will get better results . The disadvantage of the local tone mapping method is that it requires a large amount of calculation and is prone to produce a "halo" effect. In 1993, a local method was first proposed by Chiu et al. This method obtains the local brightness variation coefficient through the sensitivity of the human visual system to brightness changes, but this method cannot achieve ideal results in extremely bright and extremely dark areas. In 2002, Reinhard et al. proposed a local method for adaptively adjusting brightness based on the photographic model. This method runs faster and realizes the controllable operation of color, contrast and global brightness, but it is easy to cause halo or loss of details. . In 2002, Fattal et al. proposed a local method based on brightness gradient domain compression, which performs a dynamic range compression operation on the larger brightness gradient area of the image by adjusting the gradient attenuation function. In 2011, SylvainParis et al. applied the local Laplacian filter to HDR image dynamic range compression, and proposed an edge-aware tone mapping method for local Laplacian filtering. In 2016, Li et al. proposed a local method based on multi-scale image decomposition and guided filtering, which can better maintain global contrast and local contrast while retaining more detailed information. During the operation of this method, the image will be divided into different layers according to the actual needs, and the related processing of the image will be completed during the layering process, but because the process cannot be accurately grasped, it will cause halo phenomenon when the image is synthesized.

Contents of the invention

The purpose of the present invention is to provide a multi-scale morphology-based HDR image tone mapping method to overcome or at least alleviate at least one of the above-mentioned defects of the prior art, which realizes effective compression of the dynamic range of the HDR image and avoids halo or other artifacts, and this method has a small amount of calculation, which can meet some real-time application requirements.

In order to achieve the above object, the present invention provides a multi-scale morphology based HDR image tone mapping method, the method includes the following steps:

S1, input the HDR image to be compressed;

S2, constructing a logarithmic image of the HDR image to be compressed;

S3, performing dynamic range compression multi-scale decomposition on the logarithmic image with a first preset multiple to obtain a first Gaussian pyramid and a first Laplacian pyramid;

S4, using a one-dimensional boundary-preserving polishing operator to extract the enhanced detail layer of the first Gaussian pyramid, and adding the detailed layer of the first Gaussian pyramid to the first drawing layer by layer with a second preset multiple The layer corresponding to the Placian pyramid is obtained the second Laplacian pyramid;

S5. Perform dynamic range compression and reconstruction on the second Laplacian pyramid with a third preset multiple to obtain a detail-enhanced image after dynamic range compression;

S6. Perform exponential transformation on the detail-enhanced image to obtain a first LDR image; and

S7, performing color correction on the first LDR image to obtain an LDR image to be output;

Wherein, the method of "using a one-dimensional boundary-preserving polishing operator to extract the detail layer enhanced by the first Gaussian pyramid" in S4 specifically includes:

S41. Decompose the boundary-preserving two-dimensional polishing operator according to the horizontal and vertical directions of the HDR image to be compressed to obtain a horizontal one-dimensional boundary-preserving polishing operator and a vertical one-dimensional boundary-preserving polishing operator. light operator;

S42. According to the type of the HDR image to be compressed in S1, use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator obtained in S41 to separately perform the first Gaussian The pyramid is subjected to top-hat transformation and bottom-hat transformation layer by layer to obtain the light detail layer and dark detail layer of the first Gaussian pyramid; and

S43. Subtracting the gamma transformation result of the dark detail layer obtained in S42 layer by layer from the gamma transformation result of the bright detail layer obtained in S42, to obtain the detail layer enhanced by the first Gaussian pyramid.

Further, in S4, in the case that the HDR image to be compressed is an HDR image of a natural scene, S42 specifically includes:

S421. Use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator to perform top-hat transformation on the first Gaussian pyramid layer by layer, and convert the horizontal top-hat transformation results and The vertical top-hat transformation result is taken point by point to get the bright detail layer;

S422. Use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator to respectively perform bottom hat transformation on the first Gaussian pyramid layer by layer, and convert the results of the horizontal bottom hat transformation and The bottom hat transformation result in the vertical direction is taken point by point to get the dark detail layer.

Further, in S4, in the case that the HDR image to be compressed is a CT-HDR image, S42 specifically includes:

S421. Use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator to perform top-hat transformation on the first Gaussian pyramid layer by layer, and convert the horizontal top-hat transformation results and The vertical top-hat transformation result is taken point by point to obtain a bright detail layer;

S422. Use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator to respectively perform bottom hat transformation on the first Gaussian pyramid layer by layer, and convert the results of the horizontal bottom hat transformation and The bottom hat transformation result in the vertical direction is taken point by point to obtain the dark detail layer.

Further, S3 includes: using the following formula (1), starting from the bottom layer of the first Gaussian pyramid, performing dynamic range compression multi-scale decomposition on the first Gaussian pyramid with a first preset multiple;

In formula (1), I is the logarithmic image, {G ₀ , G ₁ ,..., G _N-1 } is the first Gaussian pyramid, and G ₀ is the maximum value of the first Gaussian pyramid. The bottom layer, G1 is the _l +1th layer of the first Gaussian pyramid, N is the number of layers of the first Gaussian pyramid, downsample represents a filter downsampling operator, and β1 is the _first preset multiple.

Further, S5 includes: using the following formula (2), starting from the topmost layer of the second Laplacian pyramid, performing dynamic range compression reconstruction on the second Laplacian pyramid with a third preset multiple , the obtained detail enhanced image G′ ₀ ;

In the formula (2), L' _N-1 is the topmost layer of the second Laplacian pyramid, G _N-1 is the top layer image of the first Gaussian pyramid, and G' _l is the middle Gaussian pyramid generated The l+1th layer, L _l is the l+1th layer of the second Laplace pyramid, is the result of filtering and upsampling the G' _l+1 layer, β ₂ is the third preset multiple, l is the l+1th layer of the second Laplacian pyramid, and N is the decomposed obtained The number of layers of the second Laplacian pyramid.

Further, the number of multi-scale decomposition layers of the first Gaussian pyramid and the first Laplacian pyramid in S3 is related to the size of the HDR image to be compressed.

Further, the second preset multiple and the third preset multiple described in S4 and S5 are related to the dynamic range of the input image, and are set to [0.35, 1].

Further, the first preset multiple described in S3 is set to [0.1, 1].

Because the present invention adopts the above technical solution, it has the following advantages: 1. The present invention realizes the effective compression of the dynamic range of HDR images. In addition, the method has a fast calculation speed and can meet certain real-time application requirements. 2. The present invention can effectively avoid the generation of halos and other artifacts during the dynamic range compression process; 3. The method of the present invention is applicable to different types of images, such as HDR images of real scenes and industrial CT images.

Description of drawings

Figure 1 is the HDR image of the input natural scene;

Fig. 2 is the image compressed by the multi-scale morphological dynamic range of Fig. 1 by using the method of the present invention;

Figure 3 is the input industrial CT-HDR image;

Fig. 4 is the compressed image of multi-scale morphological dynamic range in Fig. 2 by using the method of the present invention.

Detailed ways

In the drawings, the same or similar reference numerals are used to denote the same or similar elements or elements having the same or similar functions. Embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

The multi-scale morphology-based HDR image tone mapping method provided in this embodiment includes the following steps:

S1, input the HDR image to be compressed. Wherein, the HDR image may be the HDR image of the natural scene shown in FIG. 1 , the industrial CT-HDR image shown in FIG. 3 , or other types of HDR images.

S2. Construct a logarithmic image of the HDR image I to be compressed. Wherein, "constructing the logarithmic image of the HDR image" specifically includes: first, calculating the luminance component image of the HDR image to be compressed input by S1, and then calculating the natural logarithm of the luminance image pixel by pixel to obtain the HDR image to be compressed Log graph of I.

S3, perform dynamic range compression multi-scale decomposition on the logarithmic image with the first preset multiple to obtain the first Gaussian pyramid {G ₀ , G ₁ ,...,G _N-1 }, and then pass the first Gaussian pyramid {G ₀ , G ₁ , ..., G _N-1 } Obtain the first Laplacian pyramid {L ₀ , L ₁ , ..., L _N-1 }. Wherein, the specific implementation method of "performing dynamic range compression and multi-scale decomposition on the logarithmic image with a first preset multiple to obtain a first Gaussian pyramid" will be described in detail below. And "obtain the first Laplacian pyramid {L ₀ , L ₁ , ..., L _N-1 } through the first Gaussian pyramid {G ₀ , G ₁ , ..., G _N- 1 }" specific The implementation method is the prior art, and will not be described here. The first preset multiple is related to the dynamic range of the HDR image to be compressed input by S1, and is set to [0.35, 1].

S4, using a one-dimensional boundary-preserving polishing operator to extract the first Gaussian pyramid {G ₀ , G ₁ , ..., G _N-1 } enhanced detail layer, and the obtained first Gaussian pyramid {G ₀ , G ₁ ,...,G _N-1 } enhanced detail layers are added to the first Laplacian pyramid {L ₀ , L ₁ ,...,L _{N- 1} } corresponding layer, get the second Laplacian pyramid {L′ ₀ , L′ ₁ ,..., L′ _N-1 }. Wherein, the value range of the second preset multiple is set to [0.1, 1], and the detail enhancement is stronger. The specific implementation method of S4 will be implemented in different ways according to the type of the HDR image to be compressed in S1, and these methods will be described below.

S5. Perform dynamic range compression and reconstruction on the second Laplacian pyramid with a third preset multiple to obtain a detail-enhanced image after dynamic range compression. Wherein, the third preset multiple is related to the dynamic range of the HDR image to be compressed in S1, and experimental data shows that the dynamic range compression effect is better when set to a value within the range of [0.35,1]. The “dynamic range compression and reconstruction of the second Laplacian pyramid with a third preset multiple” will be described below.

S6. Perform exponential transformation on the detail-enhanced image to obtain a first LDR image I′. Among them, "exponential transformation" is a prior art, and will not be further described here.

S7. Perform color correction on the first LDR image to obtain an LDR image I" to be output. Wherein, "color correction" is a prior art and will not be described here.

This embodiment utilizes the above multi-scale dynamic range compression and detail enhancement operations to effectively avoid the generation of halos, and the method provided by this embodiment can be applied to effective dynamic range compression of different image types.

In one embodiment, the HDR image to be compressed inputted by S1 is a two-dimensional image. If a two-dimensional boundary-preserving smoothing operator is directly used, halos will be introduced. Therefore, in this embodiment, the two-dimensional boundary-preserving smoothing operator Decompose into two one-dimensional smoothing operators, and then use one-dimensional operators to operate on the input two-dimensional image. The one-dimensional smoothing operator in this embodiment adopts one-dimensional morphological operation, of course, one-dimensional gradient zero-norm smoothing, one-dimensional median filtering and other boundary-preserving one-dimensional smoothing operators can also be used. In view of this, S4 specifically includes:

S41. Decompose the boundary-preserving two-dimensional polishing operator according to the horizontal and vertical directions of the HDR image to be compressed to obtain a horizontal one-dimensional boundary-preserving polishing operator and a vertical one-dimensional boundary-preserving polishing operator. light operator. Among them, usually the size of the one-dimensional boundary-preserving polishing operator is 3 elements.

S42, according to the type of the HDR image to be compressed in S1, the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator respectively perform the first Gaussian pyramid {G ₀ , G ₁ ,...,G _N-1 } perform top-hat transformation and bottom-hat transformation layer by layer to obtain the bright detail layer and dark detail of the first Gaussian pyramid {G ₀ ,G ₁ ,...,G _N-1 } Floor.

S43, subtracting the gamma transformation result of the dark detail layer obtained in S42 layer by layer from the gamma transformation result of the bright detail layer obtained in S42, to obtain the first Gaussian pyramid {G ₀ , G ₁ , ..., G _N-1 } Enhanced detail layer. For example: the bright details obtained by S42 are denoted as The dark details obtained by S42 are denoted as The enhanced detail layer of the first Gaussian pyramid {G ₀ , G ₁ ,..., G _N-1 } is denoted as detail, which can be expressed as the following formula:

S44, using the following formula (3), add the obtained first Gaussian pyramid {G ₀ , G ₁ , ..., G _N-1 } enhanced detail layer details to the second preset multiple λ layer by layer Describe the layers corresponding to the first Laplacian pyramid {L ₀ , L ₁ , ..., L _N-1 }, and get the second Laplacian pyramid {L′ ₀ , L′ ₁ , ..., L ' _N-1 }.

It should be noted that if the input HDR image has a two-dimensional structure, the above "horizontal" direction refers to the row direction of the image matrix, and the "vertical" direction refers to the column direction of the image matrix.

In one embodiment, when the HDR image to be compressed in S4 is an HDR image of a natural scene, S42 specifically includes:

S421. Use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator to perform top-hat transformation on the first Gaussian pyramid layer by layer, and convert the horizontal top-hat transformation results and The vertical top-hat transformation result is taken point by point to get the bright detail layer;

S422. Use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator to respectively perform bottom hat transformation on the first Gaussian pyramid layer by layer, and convert the results of the horizontal bottom hat transformation and The bottom hat transformation result in the vertical direction is taken point by point to get the dark detail layer.

For the HDR image of the natural scene in this embodiment, the dynamic range compression result should be consistent with the perception of the human eye system. This embodiment tries to avoid the generation of halo while enhancing the details. The details in the two directions of the vertical direction are reduced, so that the blurred halo in the HDR image of the natural scene can be eliminated, that is, the halo in the HDR image of the natural scene can be eliminated, so that it is more in line with the human eye system for the real natural scene. visual perception.

In one embodiment, when the HDR image to be compressed in S4 is a CT-HDR image, S42 specifically includes:

S421. Use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator to perform top-hat transformation on the first Gaussian pyramid layer by layer, and convert the horizontal top-hat transformation results and The vertical top-hat transformation result is taken point by point to obtain a bright detail layer;

S422. Use the horizontal one-dimensional boundary-preserving polishing operator and the vertical one-dimensional boundary-preserving polishing operator to respectively perform bottom hat transformation on the first Gaussian pyramid layer by layer, and convert the results of the horizontal bottom hat transformation and The bottom hat transformation result in the vertical direction is taken point by point to obtain the dark detail layer.

The CT-HDR image aimed at in this embodiment pays more attention to the degree of preservation of image details and the quality of the overall bright and dark contrast. It is conducive to detail enhancement, and the details of the generated LDR image are more prominent.

In one embodiment, S3 includes:

Using the following formula (1), starting from the bottom layer of the first Gaussian pyramid, the dynamic range compression multi-scale decomposition is performed on the first Gaussian pyramid with a first preset multiple:

In formula (1), I is the logarithmic image, {G ₀ , G ₁ ,..., G _N-1 } is the first Gaussian pyramid, and G ₀ is the maximum value of the first Gaussian pyramid. The bottom layer, G1 is the _l +1th layer of the first Gaussian pyramid, N is the number of layers of the first Gaussian pyramid, downsample represents a filter downsampling operator, and β1 is the _first preset multiple.

In one embodiment, S5 includes:

Using the following formula (2), starting from the topmost layer of the second Laplacian pyramid, adopting a top-down manner, recursively layer by layer until l=0, for the second Laplacian The pyramid performs dynamic range compression and reconstruction at the third preset multiple, and the obtained detail enhanced image G′ ₀ is:

In formula (2), L _N-1 is the topmost layer of the second Laplacian pyramid, G _N-1 is the top layer image of the first Gaussian pyramid, and G' _l is the first layer of the intermediate Gaussian pyramid generated. l+1 layer, L _l is the l+1th layer of the second Laplace pyramid, is the result of filtering and upsampling the G' _l+1 layer, β ₂ is the third preset multiple, l is the l+1th layer of the second Laplacian pyramid, and N is the decomposed obtained The number of layers of the second Laplacian pyramid.

In one embodiment, the number of multi-scale decomposition layers of the first Gaussian pyramid and the first Laplacian pyramid in S3 is related to the size of the HDR image to be compressed. Generally, the larger the size of the HDR image, the more the number of decomposition layers. For example, an HDR image with a size of 1848*1848 can be decomposed into 8 layers.

Comparing Figure 1 and Figure 2, it is found that Figure 2 conforms to the perception of the real scene by the human eye system. Comparing Figure 3 and Figure 4, it is found that the details in Figure 4 are more prominent, which is convenient for further processing. Fig. 2 shows an HDR image in a natural scene, but the protection scope of the present invention is not limited to the dynamic range compression of the HDR image in a natural scene.

Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Those skilled in the art should understand that: the technical solutions described in the foregoing embodiments can be modified, or equivalent replacements can be made to some of the technical features; these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the various aspects of the present invention. The spirit and scope of the technical solutions of the embodiments.

Claims (8)

1. a kind of HDR image tone mapping method based on Multiscale Morphological, which comprises the steps of:

S1 inputs HDR image to be compressed；

S2 constructs the logarithmic image of the HDR image to be compressed；

S3 carries out dynamic range compression multi-resolution decomposition to the logarithmic image with the first presupposition multiple, obtains the first Gauss gold Word tower and the first laplacian pyramid；

S4, using the enhanced levels of detail of the first gaussian pyramid described in maintenance boundary polishing operator extraction, and by described the The levels of detail of one gaussian pyramid is successively added to the corresponding layer of first laplacian pyramid with the second presupposition multiple, obtains Second laplacian pyramid；

S5 carries out dynamic range compression reconstruct to second laplacian pyramid with third presupposition multiple, obtains dynamic model Enclose compressed details enhancing image；

S6 carries out exponential transform to the image of details enhancing, obtains the first LDR image；And

S7 carries out color correction to first LDR image, obtains LDR image to be output；

Wherein, in S4 " using a maintenance boundary polishing operator extraction described in the enhanced levels of detail of the first gaussian pyramid " Method specifically includes:

S41 will protect boundary two dimension polishing operator and be divided according to the horizontal and vertical both direction of the HDR image to be compressed Solution obtains horizontal maintenance boundary polishing operator and vertical maintenance boundary polishing operator；

S42 is polished according to the type of the HDR image to be compressed in S1 using the horizontal maintenance boundary that S41 is obtained Operator is polished on operator and a vertical maintenance boundary, successively carries out top cap transformation and bottom cap to first gaussian pyramid respectively Transformation, to obtain the bright levels of detail and dark levels of detail of first gaussian pyramid；With

The gamma transformation result of the obtained bright levels of detail of S42 is successively subtracted the gamma transformation for the dark levels of detail that S42 is obtained by S43 As a result, obtaining the enhanced levels of detail of the first gaussian pyramid.

2. the HDR image tone mapping method based on Multiscale Morphological as described in claim 1, which is characterized in that in S4, In the case of the HDR image to be compressed is the HDR image of natural scene, S42 is specifically included:

S421, using horizontal maintenance boundary polishing operator and vertical maintenance boundary polishing operator respectively to described first Gaussian pyramid successively carries out top cap transformation, by every layer of horizontal direction top cap transformation results and vertical direction top cap transformation results by Point takes small, obtains bright levels of detail；

S422, using horizontal maintenance boundary polishing operator and vertical maintenance boundary polishing operator respectively to described first Gaussian pyramid successively carries out bottom cap transformation, by every layer of horizontal direction bottom cap transformation results and vertical direction bottom cap transformation results by Point takes small, obtains dark levels of detail.

3. the HDR image tone mapping method based on Multiscale Morphological as described in claim 1, which is characterized in that in S4, The HDR image to be compressed is in the case of CT-HDR image, S42 is specifically included:

S421, using horizontal maintenance boundary polishing operator and vertical maintenance boundary polishing operator respectively to described first Gaussian pyramid successively carries out top cap transformation, by every layer of horizontal direction top cap transformation results and vertical direction top cap transformation results by Point takes greatly, obtains bright levels of detail；

S422, using horizontal maintenance boundary polishing operator and vertical maintenance boundary polishing operator respectively to described first Gaussian pyramid successively carries out bottom cap transformation, by every layer of horizontal direction bottom cap transformation results and vertical direction bottom cap transformation results by Point takes greatly, obtains dark levels of detail.

4. the HDR image tone mapping method based on Multiscale Morphological as claimed any one in claims 1 to 3, special Sign is that S3 includes: to utilize such as following formula (1), since the bottom of first gaussian pyramid, to the first Gauss gold Word tower carries out dynamic range compression multi-resolution decomposition with the first presupposition multiple；

In formula (1), I is the logarithmic image, { G₀, G₁... ..., G_N-1It is the first gaussian pyramid, G₀It is first Gauss The pyramidal bottom, G_lIt is l+1 layers of first gaussian pyramid, N is the number of plies of first gaussian pyramid, Downsample indicates filtering down-sampling operator, β₁For first presupposition multiple.

5. the HDR image tone mapping method based on Multiscale Morphological as claimed in claim 4, which is characterized in that S5 packet It includes: using following formula (2), since the top of second laplacian pyramid, to second laplacian pyramid Dynamic range compression reconstruct is carried out with third presupposition multiple, obtained details enhancing image G '₀；

In formula (2), L '_N-1It is the top of second laplacian pyramid, G_N-1It is the top of first gaussian pyramid Tomographic image, G '_lIt is pyramidal l+1 layers of middle Gaussian generated, L_lIt is l+1 layers of second laplacian pyramid,It is to G '_l+1Layer filtering up-sampling as a result, β₂For the third presupposition multiple, l is second laplacian pyramid L+1 layers, N is the number of plies for decomposing obtained second laplacian pyramid.

6. the HDR image tone mapping method based on Multiscale Morphological as claimed in claim 5, which is characterized in that in S3 The multi-resolution decomposition number of plies of first gaussian pyramid and the first laplacian pyramid and the HDR image to be compressed Size is related.

7. the HDR image tone mapping method based on Multiscale Morphological as described in claim 1, which is characterized in that S4 and Second presupposition multiple described in S5 and third presupposition multiple are related with the dynamic range of input picture, are set as [0.35,1].

8. the HDR image tone mapping method based on Multiscale Morphological as described in claim 1, which is characterized in that in S3 First presupposition multiple is set as [0.1,1].