WO2020248294A1 - 一种逆色调映射方法、装置及电子设备 - Google Patents

一种逆色调映射方法、装置及电子设备 Download PDF

Info

Publication number
WO2020248294A1
WO2020248294A1 PCT/CN2019/091874 CN2019091874W WO2020248294A1 WO 2020248294 A1 WO2020248294 A1 WO 2020248294A1 CN 2019091874 W CN2019091874 W CN 2019091874W WO 2020248294 A1 WO2020248294 A1 WO 2020248294A1
Authority
WO
WIPO (PCT)
Prior art keywords
dynamic range
layer
range image
low dynamic
network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/091874
Other languages
English (en)
French (fr)
Inventor
王荣刚
王超
高文
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Peking University Shenzhen Graduate School
Original Assignee
Peking University Shenzhen Graduate School
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Peking University Shenzhen Graduate School filed Critical Peking University Shenzhen Graduate School
Priority to US16/975,485 priority Critical patent/US11783459B2/en
Publication of WO2020248294A1 publication Critical patent/WO2020248294A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G06T5/94—Dynamic range modification of images or parts thereof based on local image properties, e.g. for local contrast enhancement
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G06T5/92—Dynamic range modification of images or parts thereof based on global image properties
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/60—Image enhancement or restoration using machine learning, e.g. neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20172—Image enhancement details
    • G06T2207/20192—Edge enhancement; Edge preservation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20172—Image enhancement details
    • G06T2207/20208—High dynamic range [HDR] image processing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging

Definitions

  • This specification relates to the technical field of digital image processing, and in particular to an inverse tone mapping method, device and electronic equipment.
  • inverse tone mapping inverse tone mapping
  • 4K TV technology Take the application of inverse tone mapping technology in 4K TV technology as an example. As most media resources are still stored in low dynamic range, the inverse tone mapping technology is used to convert media resources from low dynamic range to high dynamic range. An important part of 4K TV technology.
  • the purpose of the present invention is to provide an inverse tone mapping method, device and electronic equipment to solve the problem of poor conversion effect from low dynamic range image to high dynamic range image in the prior art.
  • a predetermined fusion network is used to adjust the restored detail layer and the basic layer to obtain an adjusted high dynamic range image.
  • the acquiring one or more low dynamic range images includes: compressing the original image to obtain the compressed low dynamic range image.
  • the performing a decomposition operation on the low dynamic range image to obtain the detail layer and the basic layer of the low dynamic range image includes:
  • the low dynamic range image is decomposed based on the Retinex theory to obtain the detail layer and the basic layer of the low dynamic range image.
  • the decomposition of the low dynamic range image based on the Retinex theory to obtain the detail layer and the base layer of the low dynamic range image specifically includes:
  • the first restoration network is a residual network
  • the second restoration network is a U-Net network
  • the use of a predetermined first restoration network and a second restoration network respectively performs the analysis on the detail layer and the basic Layer to restore, including:
  • the residual network is used to restore the detail layer, and the U-Net network is used to restore the basic layer.
  • the detail layer includes high-frequency components and compression artifacts of the low dynamic range image
  • the base layer includes low-frequency components of the low dynamic range image
  • the residual network Restoring the detail layer and using the U-Net network to restore the basic layer includes:
  • the residual network is used to restore the high-frequency components to remove the compression artifacts; and the U-Net network is used to restore the low-frequency components.
  • the high frequency components include edges and contours
  • the low frequency components include color information and structure information.
  • the fusion network adopts a residual network.
  • the acquisition module is used to acquire one or more low dynamic range images
  • a decomposition module configured to perform a decomposition operation on the low dynamic range image to obtain the detail layer and the basic layer of the low dynamic range image
  • the restoration module is configured to use a predetermined first restoration network and a second restoration network to restore the detail layer and the basic layer, respectively, to obtain the restored detail layer and the basic layer;
  • the adjustment module is used to adjust the restored detail layer and basic layer by using a predetermined fusion network to obtain an adjusted high dynamic range image.
  • the acquisition module is specifically configured to: compress the original image to obtain the compressed low dynamic range image.
  • the decomposition module is specifically configured to decompose the low dynamic range image based on Retinex theory to obtain the detail layer and the basic layer of the low dynamic range image.
  • the decomposition module is further configured to: perform edge-preserving filtering on the low dynamic range image, and use the image obtained after the edge-preserving filtering as the basic layer of the low dynamic range image; The difference between the dynamic range image and the base layer image is calculated, and the image obtained after the difference is used as the detail layer of the low dynamic range image.
  • the first recovery network is a residual network
  • the second recovery network is a U-Net network
  • the recovery module is specifically configured to: use the residual network to recover the detail layer
  • the U-Net network is used to restore the basic layer.
  • the detail layer includes high-frequency components and compression artifacts of the low dynamic range image
  • the base layer includes low-frequency components of the low dynamic range image
  • the restoration module is further configured to:
  • the residual network is used to restore the high-frequency components to remove the compression artifacts; and the U-Net network is used to restore the low-frequency components.
  • the high frequency components include edges and contours
  • the low frequency components include color information and structure information.
  • An electronic device provided by an embodiment of this specification includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the above-mentioned inverse tone mapping method when the program is executed.
  • the present invention obtains one or more low dynamic range images and performs a decomposition operation on the low dynamic range images to obtain the detail layer and the basic layer of the low dynamic range image; using a predetermined first restoration network and a second restoration network, respectively
  • the detail layer and the basic layer are restored to obtain the restored detail layer and the basic layer; the restored detail layer and the basic layer are adjusted using the predetermined fusion network to obtain the adjusted high dynamic range image.
  • FIG. 1 is a schematic flowchart of an inverse tone mapping method provided by an embodiment of this specification
  • Figure 2 is a schematic structural diagram of a residual network provided by an embodiment of this specification.
  • FIG. 3 is a schematic diagram of the structure of the U-Net network provided by an embodiment of this specification.
  • Figure 4 is a schematic diagram of the structure of a converged network provided by an embodiment of this specification.
  • FIG. 5 is a schematic structural diagram of an inverse tone mapping apparatus provided by an embodiment of this specification.
  • the current inverse tone mapping methods are mostly aimed at high-quality low dynamic range images, that is, for lossless compressed or uncompressed low dynamic range images.
  • the traditional inverse tone mapping method usually proposes a parametric model and completes the low dynamic range through the parametric model. Conversion of dynamic range image to high dynamic range image. This traditional method is mainly to expand the brightness, so as to display a good visual effect on the high dynamic range display. However, this method has many drawbacks. It cannot completely recover the information lost in the low dynamic range image, and it also requires complicated parameter settings, which is also very difficult for ordinary users.
  • ordinary images are stored in a lossy compressed format during transmission, which will cause compression artifacts, and compression artifacts have a serious impact on the result of inverse tone mapping.
  • low-quality low dynamic range images that is, low dynamic range images stored in a lossy compression format
  • the following embodiments of this specification may be performed on a low dynamic range image as a processing object, especially a low dynamic range image after lossy compression (ie, a low-quality low dynamic range image) as a processing object.
  • a low dynamic range image after lossy compression ie, a low-quality low dynamic range image
  • FIG. 1 is a schematic flowchart of an inverse tone mapping method provided by an embodiment of the present invention. The method may specifically include the following steps:
  • step S110 one or more low dynamic range images are acquired.
  • the low dynamic range image is an object of inverse tone mapping processing
  • acquiring one or more low dynamic range images can be regarded as acquiring one or more images to be processed.
  • the image to be processed in the embodiments of this specification can be any low dynamic range image, including but not limited to: uncompressed low dynamic range image, lossless compressed low dynamic range image, and lossy compressed low dynamic range image Range image.
  • uncompressed low dynamic range image lossless compressed low dynamic range image
  • Lossy compressed low dynamic range images also need to consider the impact of compression artifacts on inverse tone mapping.
  • a compressed low dynamic range image may be obtained by compressing the original image, where the compression may be lossy compression, and the original image may include uncompressed low dynamic range images, lossless compression Low dynamic range image or high dynamic range image, etc.
  • the following embodiments of this specification use lossy compressed low dynamic range images, that is, low quality low dynamic range images as the image to be processed.
  • a low dynamic range image can be considered as an image with a pixel value between 0-255
  • a high dynamic range image can be considered as an image with a pixel value between 0-16 ⁇ 10 5 image.
  • step S120 a decomposition operation is performed on the low dynamic range image to obtain the detail layer and the basic layer of the low dynamic range image.
  • the decomposition operation can be performed on the obtained low-dynamic-range image to obtain the low-dynamic-range image Detail layer and basic layer.
  • the following method may be used to perform the decomposition operation on the low dynamic range image, and the method may include the following process:
  • the low dynamic range image is decomposed based on Retinex theory, and the detail layer and basic layer of the low dynamic range image are obtained. Specifically, edge-preserving filtering is performed on the low dynamic range image, and the image obtained after edge-preserving filtering is used as the base layer of the low dynamic range image; and the difference between the low dynamic range image and the base layer image is calculated, The image obtained after the difference is used as the detail layer of the low dynamic range image.
  • Retinex theory is a theory widely used in digital image processing. It believes that digital images can be decomposed into detail layers and basic layers, which are independent and do not affect each other.
  • the low dynamic range image is decomposed into high-frequency and low-frequency parts by using Retinex theory, that is, decomposed into Detail layer and base layer; among them, the detail layer contains the high frequency components and compression artifacts of the low dynamic range image, and the base layer contains the low frequency components of the low dynamic range image.
  • step S130 the detailed layer and the base layer are respectively restored by using the predetermined first restoration network and the second restoration network to obtain the restored detail layer and the basic layer.
  • the high frequency components may include texture information of low dynamic range images, such as edges and contours. Wait. Therefore, for the detail layer, it is necessary to ensure the integrity of its structure, reduce the loss of information, and avoid down-sampling, so the residual network with the same feature map size can be used as the recovery network of the detail layer.
  • the basic layer contains the low-frequency component of the low dynamic range image, and the low-frequency component contains a lot of color information and structural information (such as object shape), etc. This part of information is useful for restoring overexposed areas, underexposed areas, and colors. Mapping plays an important role. In order to extract sufficient features, it is necessary to use multi-scale information to restore the object structure, so U-Net network can be used as the basic layer of restoration network.
  • the residual network to restore the detail layer
  • the U-Net network to restore the basic layer
  • the first restoration network is the residual network
  • the second restoration network is the U-Net network
  • the detail layer contains the high frequency components and compression artifacts of the low dynamic range image
  • the base layer contains the low frequency components of the low dynamic range image
  • the residual network is used to restore the detail layer and use U-Net network restores the basic layer, which can include the following:
  • the residual network is used to recover high-frequency components to remove compression artifacts; and the U-Net network is used to recover low-frequency components.
  • the residual network includes one or more convolutional layers on both sides and a plurality of residual blocks in the middle, and each residual block contains a first convolutional layer, a second activation layer, and a third volume in order.
  • a build-up layer and a fourth activation layer wherein, before the fourth activation layer, the method further includes: performing an addition operation on the input image of the residual block and the output image of the third convolutional layer.
  • the residual network may include two convolutional layers on the front and back sides and 16 residual blocks in the middle.
  • the activation layer in the residual block adopts the SELU activation function.
  • the size of the convolution kernel of each convolutional layer is 3*3, and the step size is 1. Except for the feature channel number of the last convolutional layer (that is, the rightmost convolutional layer in Figure 2), the number of feature channels is 3.
  • the number of characteristic channels is 64.
  • a mirror image symmetry method can be used for edge filling.
  • the U-Net network includes multiple convolution blocks and deconvolution blocks.
  • the convolution block is located in front of the deconvolution block.
  • the convolution block includes a convolution layer, an activation layer, and a convolution layer in order.
  • the activation layer in order to avoid checkerboard artifacts, in the deconvolution block, first upsample, expand the resolution of the feature map, and then perform the convolution operation.
  • the deconvolution block contains the upsampling layer and convolution in order. Layer and active layer. At the same time, in order to speed up the convergence speed, each layer adds batch normalization operation, and in order to make full use of low-level features, there are jump links between the corresponding upper and lower sampling layers.
  • the U-Net network includes 5 convolution blocks and 4 deconvolution blocks, that is, the U-Net network can include a total of 9 layers; among them, the first four convolution blocks, Each convolutional block sequentially contains a convolutional layer with a size of 1*1, a step size of 1, an activation layer, a convolutional layer with a size of 3*3, a step size of 2, and an activation layer, the first four volumes
  • the number of characteristic channels of the block is 64, 128, 256, 512 respectively;
  • the fifth convolution block contains in order the convolutional layer with size 3*3 and step size 1, and the activation layer, the size is 3*3 ,
  • the convolutional layer and the activation layer with a step size of 1, the number of characteristic channels of the fifth convolutional block is 1024.
  • the deconvolution block contains an up-sampling layer in order, a convolution layer with a size of 3*3, a step size of 1, and an activation layer.
  • the number of characteristic channels of the deconvolution block is 512, 256, 128, 64, 3, respectively .
  • the up-sampling is double-line up-sampling, and the resolution of the feature map can be enlarged through up-sampling.
  • step S140 a predetermined fusion network is used to adjust the restored detail layer and base layer to obtain an adjusted high dynamic range image.
  • the restored detail layer and the basic layer can be obtained, and finally the restored detail layer and the The basic layer is further adjusted to obtain the final restored high dynamic range image.
  • the fusion network may adopt a residual network, and the structure of the fusion network is similar to that of the first restoration network (ie, the detail layer restoration network), but in practical applications, the fusion network may only include 8 residual blocks.
  • Fig. 4 shows a schematic structural diagram of a converged network provided by an embodiment of this specification. Since both the converged network and the first recovery network adopt residual networks and their structures are similar, the structure of the converged network will not be repeated here.
  • the structure of the converged network please refer to the above description of the first recovery network structure.
  • the low dynamic range image is decomposed into high-frequency parts (ie, the detail layer) with the help of Retinex theory. ) And the low frequency part (ie the basic layer).
  • the detail layer contains the high-frequency components (such as edges and contours) and compression artifacts of the image, it is mainly responsible for the preservation of edges and the removal of compression artifacts; while the basic layer contains the low-frequency components of the image (such as color information, structure information), Mainly responsible for the restoration of overexposed and underexposed missing information and color mapping, so for the above two different component information, two different networks are used to restore the detail layer and the basic layer, and finally a fusion network is used The results of the first two networks are further adjusted to achieve the restoration of high dynamic range images.
  • the present invention can not only recover the information lost in the low dynamic range image, but also take into account the recovery of different missing information, such as: overexposed area, underexposed area, color information and object structure, etc., but can also remove compression artifacts at the same time, making it more robust Complete the conversion of compressed low dynamic range images to high dynamic range images.
  • the embodiment of this specification also provides an inverse tone mapping device, as shown in Fig. 5 is an inverse tone mapping device provided by the embodiment of this specification.
  • the device 500 mainly includes:
  • the obtaining module 501 is used to obtain one or more low dynamic range images
  • the decomposition module 502 is configured to perform a decomposition operation on the low dynamic range image to obtain the detail layer and the basic layer of the low dynamic range image;
  • the restoration module 503 is configured to use a predetermined first restoration network and a second restoration network to respectively restore the detail layer and the basic layer to obtain the restored detail layer and the basic layer;
  • the adjustment module 504 is configured to adjust the restored detail layer and the basic layer by using a predetermined fusion network to obtain an adjusted high dynamic range image.
  • the acquisition module 501 is specifically configured to: compress the original image to obtain the compressed low dynamic range image.
  • the decomposition module 502 is specifically configured to decompose the low dynamic range image based on the Retinex theory to obtain the detail layer and the base layer of the low dynamic range image.
  • the decomposition module 502 is further configured to: perform edge preserving filtering on the low dynamic range image, and use the image obtained after the edge preserving filtering as the low dynamic range image The base layer of the image; and the difference between the low dynamic range image and the base layer image is calculated, and the image obtained after the difference is used as the detail layer of the low dynamic range image.
  • the first recovery network is a residual network
  • the second recovery network is a U-Net network
  • the recovery module 503 is specifically configured to: use the residual The network restores the detail layer, and uses the U-Net network to restore the basic layer.
  • the detail layer includes high frequency components and compression artifacts of the low dynamic range image
  • the base layer includes low frequency components of the low dynamic range image
  • the recovery module 503 is further used to:
  • the residual network is used to restore the high-frequency components to remove the compression artifacts; and the U-Net network is used to restore the low-frequency components.
  • the high frequency components include edges and contours
  • the low frequency components include color information and structure information.
  • the embodiment of the present specification also provides an electronic device including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the above-mentioned inverse tone mapping method when the program is executed.
  • the device, electronic device, and method provided in the embodiments of this specification are corresponding. Therefore, the device and electronic device also have beneficial technical effects similar to the corresponding method. Since the beneficial technical effects of the method have been described in detail above, here The beneficial technical effects of the corresponding devices and electronic equipment will not be repeated.
  • program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types.
  • the instructions can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network.
  • program modules can be located in local and remote computer storage media including storage devices.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)
  • Compression Of Band Width Or Redundancy In Fax (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

本说明书实施例提供一种逆色调映射方法、装置及电子设备。所述方法包括：获取一个或多个低动态范围图像；对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层；利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，得到所述恢复后的细节层和基本层；利用预定的融合网络对所述恢复后的细节层和基本层进行调整，得到调整后的高动态范围图像。采用本申请的技术方案，能够更加鲁棒地完成低动态范围图像向高动态范围图像的转换。

Description

一种逆色调映射方法、装置及电子设备
本申请要求享有2019年6月10日提交的名称为“一种逆色调映射方法、装置及电子设备”的中国专利申请CN201910499995.3的优先权，其全部内容通过引用并入本文中。
技术领域
本说明书涉及数字图像处理技术领域，尤其涉及一种逆色调映射方法、装置及电子设备。
背景技术
在数字图像处理领域，普通图像若要在高动态范围(High Dynamic Range，HDR)显示设备上显示则无法提供足够的精度，所以需要从普通图像恢复HDR信息，这一过程称为逆色调映射(inverse tone mapping)。以逆色调映射技术在4K电视技术的应用为例，由于目前大多数媒体资源仍为低动态范围的存储方式，所以利用逆色调映射技术将媒体资源从低动态范围转换到高动态范围，这是4K电视技术中的重要部分。
现有技术中，通过提出一个参数模型，通过参数模型完成低动态范围图像到高动态范围图像的转换，这种方式主要是对亮度的扩展，以便于在高动态范围显示器上显示出良好的视觉效果。但是现有的这种方式无法完全恢复低动态范围图像中损失的信息，而且需要复杂的参数设置。另外，目前的逆色调映射方法大多针对高质量的低动态范围图像，然而实际中，很多媒体资源在传输的过程中被存储为有损压缩的格式，这将会导致压缩伪影的产生，而压缩伪影对逆色调映射的结果有严重影响。
基于现有技术，需要提供一种能鲁棒地完成低动态范围图像向高动态范围 图像转换的逆色调映射方案。
发明内容
有鉴于此，本发明的目的在于提供一种逆色调映射方法、装置及电子设备，以解决现有技术存在的低动态范围图像向高动态范围图像的转换效果差的问题。
为解决上述技术问题，本说明书实施例是这样实现的：
本说明书实施例提供的一种逆色调映射方法，所述方法包括：
获取一个或多个低动态范围图像；
对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层；
利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，得到所述恢复后的细节层和基本层；
利用预定的融合网络对所述恢复后的细节层和基本层进行调整，得到调整后的高动态范围图像。
可选的，所述获取一个或多个低动态范围图像，包括：对原始图像进行压缩，得到所述压缩后的低动态范围图像。
可选的，所述对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层，包括：
基于Retinex理论对所述低动态范围图像进行分解，得到所述低动态范围图像的细节层和基本层。
可选的，所述基于Retinex理论对所述低动态范围图像进行分解，得到所述低动态范围图像的细节层和基本层，具体包括：
对所述低动态范围图像进行保边滤波，将所述保边滤波后得到的图像作为所述低动态范围图像的基本层；并对所述低动态范围图像与基本层图像进行求 差，将所述求差后得到的图像作为所述低动态范围图像的细节层。
可选的，所述第一恢复网络为残差网络，所述第二恢复网络为U-Net网络；所述利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，包括：
利用所述残差网络对所述细节层进行恢复，并利用所述U-Net网络对所述基本层进行恢复。
可选的，所述细节层包含所述低动态范围图像的高频分量和压缩伪影，所述基本层包含所述低动态范围图像的低频分量；所述利用所述残差网络对所述细节层进行恢复，并利用所述U-Net网络对所述基本层进行恢复，包括：
利用所述残差网络对所述高频分量进行恢复，以便去除所述压缩伪影；并利用所述U-Net网络对所述低频分量进行恢复。
可选的，所述高频分量包括边缘和轮廓，所述低频分量包括颜色信息和结构信息。
可选的，所述融合网络采用残差网络。
本说明书实施例提供的一种逆色调映射装置，所述装置包括：
获取模块，用于获取一个或多个低动态范围图像；
分解模块，用于对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层；
恢复模块，用于利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，得到所述恢复后的细节层和基本层；
调整模块，用于利用预定的融合网络对所述恢复后的细节层和基本层进行调整，得到调整后的高动态范围图像。
可选的，所述获取模块具体用于：对原始图像进行压缩，得到所述压缩后的低动态范围图像。
可选的，所述分解模块具体用于：基于Retinex理论对所述低动态范围图像 进行分解，得到所述低动态范围图像的细节层和基本层。
可选的，所述分解模块进一步用于：对所述低动态范围图像进行保边滤波，将所述保边滤波后得到的图像作为所述低动态范围图像的基本层；并对所述低动态范围图像与基本层图像进行求差，将所述求差后得到的图像作为所述低动态范围图像的细节层。
可选的，所述第一恢复网络为残差网络，所述第二恢复网络为U-Net网络；所述恢复模块具体用于：利用所述残差网络对所述细节层进行恢复，并利用所述U-Net网络对所述基本层进行恢复。
可选的，所述细节层包含所述低动态范围图像的高频分量和压缩伪影，所述基本层包含所述低动态范围图像的低频分量；所述恢复模块进一步用于：
利用所述残差网络对所述高频分量进行恢复，以便去除所述压缩伪影；并利用所述U-Net网络对所述低频分量进行恢复。
可选的，所述高频分量包括边缘和轮廓，所述低频分量包括颜色信息和结构信息。
本说明书实施例提供的一种电子设备，包括存储器，处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现上述一种逆色调映射方法。
本说明书实施例采用的上述至少一个技术方案能够达到以下有益效果：
本发明通过获取一个或多个低动态范围图像，并对低动态范围图像执行分解操作，得到低动态范围图像的细节层和基本层；利用预定的第一恢复网络和第二恢复网络，分别对细节层和基本层进行恢复，得到恢复后的细节层和基本层；利用预定的融合网络对恢复后的细节层和基本层进行调整，得到调整后的高动态范围图像。采用本申请的技术方案，能够更加鲁棒地完成低动态范围图像向高动态范围图像的转换，并且无需复杂的参数设置。
附图说明
为了更清楚的说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单的介绍，显而易见的下面描述中的附图仅仅是本发明的实施例，对于本领域普通人员来讲，在不付出创造性劳动的前提下，还可以根据提供的附图获得其他的附图。
图1是本说明书实施例提供的一种逆色调映射方法的流程示意图；
图2是本说明书实施例提供的残差网络的结构示意图；
图3是本说明书实施例提供的U-Net网络的结构示意图；
图4是本说明书实施例提供的融合网络的结构示意图；
图5是本说明书实施例提供的一种逆色调映射装置的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本说明书中的技术方案，下面将结合本说明书实施例中的附图，对本说明书实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本说明书实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都应当属于本申请保护的范围。
在实际应用中，由于SDR(Standard Dynamic Range)内容在HDR显示设备上显示无法提供足够的精度，而且当前的HDR内容短缺，因此亟需一种能将SDR内容转换为HDR内容的技术，逆色调映射技术应运而生。目前大多数媒体资源(如图像)仍为低动态范围的存储方式，并且为了便于将图像进行存储和传输，需要将普通图像以有损压缩的格式进行存储，而这也会引入压缩伪影，如块状伪影、带状伪影和振铃效应等。
目前的逆色调映射方法大多针对高质量的低动态范围图像，即针对的是无损压缩或未经压缩的低动态范围图像，传统的逆色调映射方法通常是提出一个 参数模型，通过参数模型完成低动态范围图像到高动态范围图像的转换。这种传统方式主要是对亮度的扩展，以便于在高动态范围显示器上显示出良好的视觉效果。但是这种方式存在较多弊端，无法完全恢复低动态范围图像中损失的信息，而且还需要复杂的参数设置，对普通用户来说也很困难。除此之外，普通图像在传输的过程中以有损压缩的格式进行存储，这会导致压缩伪影的产生，而压缩伪影对逆色调映射的结果有严重影响。如果先去除压缩伪影再使用逆色调映射会导致结果过于模糊，如果在逆色调映射的基础上再去除压缩伪影，伪影会因为逆色调映射而被增强难以去除。
因此，针对低质量的低动态范围图像，即以有损压缩格式存储的低动态范围图像，需要提供一种既能恢复低动态范围图像在压缩过程中损失的信息，又能同时去除压缩伪影，更加鲁棒地完成低动态范围图像向高动态范围图像转换的逆色调映射方案。本说明书以下实施例可以是针对低动态范围图像作为处理对象进行的，特别是针对有损压缩后的低动态范围图像(即低质量的低动态范围图像)作为处理对象而进行的。当然，以有损压缩后的低动态范围图像作为处理对象仅仅是本说明书实施例的一种应用场景，不能构成对本说明书实施例应用范围的限定。
图1为本发明实施例提供的一种逆色调映射方法的流程示意图。该方法具体可以包括以下步骤：
在步骤S110中，获取一个或多个低动态范围图像。
在本说明书一个或多个实施例中，所述低动态范围图像是逆色调映射处理的对象，获取一个或多个低动态范围图像，即可以认为是获取一个或多个待处理图像。根据前述内容，本说明书实施例中的待处理图像可以是任何的低动态范围图像，包括但不限于：未经压缩的低动态范围图像、无损压缩的低动态范围图像和有损压缩的低动态范围图像。考虑到现有的大多数媒体资源为了便于存储和传输，因此均被处理为有损压缩的格式图像，而有损压缩的方式导致了压缩伪影的产生，所以相比其它的低动态范围图像，有损压缩的低动态范围图 像还需要考虑压缩伪影对逆色调映射的影响。
在一个具体实施例中，可以通过对原始图像进行压缩，得到压缩后的低动态范围图像，这里的压缩可以是有损压缩，所述原始图像可以包括未经压缩的低动态范围图像、无损压缩的低动态范围图像或者高动态范围图像等。优选地，本说明书以下实施例是以有损压缩的低动态范围图像，即将低质量的低动态范围图像作为待处理图像而说的。
值得说明的是，在本说明书实施例中，低动态范围图像可以认为是像素值在0-255之间的图像，而高动态范围图像可以认为是像素值在0-16×10 5之间的图像。
在步骤S120中，对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层。
在本说明书一个或多个实施例中，继续上述实施例的内容，在经过压缩处理获得低质量的低动态范围图像之后，可以对获得的低动态范围图像执行分解操作，得到低动态范围图像的细节层和基本层。本说明书实施例中，可以采用以下方法对低动态范围图像执行分解操作，该方法可以包括以下过程：
基于Retinex理论对低动态范围图像进行分解，得到低动态范围图像的细节层和基本层。具体地，对所述低动态范围图像进行保边滤波，将保边滤波后得到的图像作为低动态范围图像的基本层；并对所述低动态范围图像与基本层图像进行求差，将求差后得到的图像作为低动态范围图像的细节层。
Retinex理论是一种广泛应用于数字图像处理中的理论，它认为数字图像可以被分解为细节层和基本层，两者独立互不影响。
进一步地，考虑到压缩伪影是由于对高频分量的过度压缩导致的，主要存在于高频分量部分，因此通过采用Retinex理论将低动态范围图像分解为高频部分和低频部分，即分解为细节层和基本层；其中，细节层包含了低动态范围图像的高频分量和压缩伪影，基本层包含了低动态范围图像的低频分量。
在步骤S130中，利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，得到所述恢复后的细节层和基本层。
在本说明书一个或多个实施例中，由于细节层包含了低动态范围图像的高频分量和压缩伪影，其中所述高频分量又可以包括低动态范围图像的纹理信息，如边缘和轮廓等。因此，对于细节层需要保证其结构的完整性，减少信息的损失，避免下采样，所以可以采用特征图尺寸不变的残差网络作为细节层的恢复网络。
针对基本层而言，其包含了低动态范围图像的低频分量，而低频分量中含有大量的颜色信息和结构信息(如物体形状)等，这部分信息对于恢复过曝区域、欠曝区域和颜色映射具有重要作用，为了提取充足的特征，因此需要利用多尺度信息来恢复物体结构，所以可以采用U-Net网络作为基本层的恢复网络。
进一步地，利用所述残差网络对细节层进行恢复，并利用所述U-Net网络对基本层进行恢复；此时，第一恢复网络为残差网络，第二恢复网络为U-Net网络。
根据前述实施例的内容，细节层包含了低动态范围图像的高频分量和压缩伪影，基本层包含了低动态范围图像的低频分量；因此，利用残差网络对细节层进行恢复，并利用U-Net网络对基本层进行恢复，可以包括以下内容：
利用所述残差网络对高频分量进行恢复，以便去除压缩伪影；并利用所述U-Net网络对低频分量进行恢复。
下面结合附图对残差网络和U-Net网络的结构进行详细地描述，如图2-图3所示，为本说明书实施例提供的残差网络和U-Net网络的结构示意图，具体可以包括以下内容：
残差网络包括两侧的一个或多个卷积层以及中间的多个残差块，每个所述残差块内包含按顺序排列的第一卷积层、第二激活层、第三卷积层和第四激活层，其中，在所述第四激活层之前还包括：对所述残差块的输入图像与所述第三卷积层的输出图像执行相加操作。
进一步地，在本说明书一个实施例中，残差网络可以包括前后两侧的2层卷积层和中间的16个残差块，残差块内的激活层采用SELU激活函数，残差网络中各卷积层的卷积核尺寸为3*3，步长为1，除最后一层卷积层(即图2中最右侧的卷积层)的特征通道数为3外，其余各个层的特征通道数均为64。在实际应用中，为了保持特征图尺寸不变可以采用镜像对称的方式做边缘填充。
U-Net网络包括多个卷积块和反卷积块，卷积块位于反卷积块的前面，其中，所述卷积块内包含按顺序排列的卷积层、激活层、卷积层以及激活层；为了避免棋盘伪影，在反卷积块中，先上采样，扩大特征图的分辨率，再进行卷积操作，反卷积块内包含按顺序排列的上采样层、卷积层以及激活层。同时为了加快收敛速度，每一层都加入批归一化操作，并且为了充分利用低级特征，对应的上下采样层之间有跳转链接。
进一步地，在本说明书一个实施例中，U-Net网络包括5个卷积块和4个反卷积块，即U-Net网络内一共可以包括9层；其中，前四个卷积块，每个卷积块内按顺序依次包含尺寸为1*1、步长为1的卷积层，激活层，尺寸为3*3、步长为2的卷积层以及激活层，前四个卷积块的特征通道数目分别为64、128、256、512；第五个卷积块内按顺序依次包含尺寸为3*3、步长为1的卷积层，激活层，尺寸为3*3、步长为1的卷积层以及激活层，第五个卷积块的特征通道数目为1024。反卷积块内按顺序依次包含上采样层，尺寸为3*3、步长为1的卷积层以及激活层，反卷积块的特征通道数分别为512、256、128、64、3。
在本说明书实施例中，所述上采样采用的是双线向上采样，通过上采样可以扩大特征图的分辨率。
在步骤S140中，利用预定的融合网络对所述恢复后的细节层和基本层进行调整，得到调整后的高动态范围图像。
继续上述实施例，在利用两个不同的恢复网络对高频分量和低频分量分别进行恢复后，便可以得到恢复后的细节层和基本层，最后利用预定的融合网络将恢复后的细节层和基本层做进一步调整，得到最终恢复后的高动态范围图像。
进一步地，所述融合网络可以采用残差网络，融合网络的结构与第一恢复网络(即细节层恢复网络)的结构相似，但是在实际应用中，融合网络可以只包含8个残差块。参见图4，该图示出了本说明书实施例提供的融合网络的结构示意图。由于融合网络与第一恢复网络均采用残差网络，且两者的结构相似，因此这里不再对融合网络的结构进行赘述，关于融合网络的结构可以参见上述对第一恢复网络结构的描述。
基于上述本说明书实施例，考虑到压缩伪影是由于对高频分量的过度压缩导致的，主要存在于高频分量部分，因此借助Retinex理论将低动态范围图像分解为高频部分(即细节层)和低频部分(即基本层)。由于细节层包含图像的高频分量(如边缘、轮廓)和压缩伪影，主要负责边缘的保持和压缩伪影的去除；而基本层则包含图像的低频分量(如颜色信息、结构信息)，主要负责对过曝和欠曝缺失信息的恢复以及颜色的映射，因此针对上述两种不同的分量信息，分别采用两个不同的网络对细节层和基本层进行恢复，最后再利用一支融合网络将前两支网络的结果进一步调整，从而实现高动态范围图像的恢复。本发明不仅能恢复低动态范围图像中损失的信息，兼顾不同缺失信息的恢复，例如：过曝区域、欠曝区域、颜色信息和物体结构等，还可以同时去除压缩伪影，更加鲁棒地完成压缩低动态范围图像到高动态范围图像的转换。
基于同样的思路，本说明书实施例还提供了一种逆色调映射装置，如图5为本说明书实施例提供的一种逆色调映射装置，该装置500主要包括：
获取模块501，用于获取一个或多个低动态范围图像；
分解模块502，用于对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层；
恢复模块503，用于利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，得到所述恢复后的细节层和基本层；
调整模块504，用于利用预定的融合网络对所述恢复后的细节层和基本层进行调整，得到调整后的高动态范围图像。
根据本申请的实施例，在所述装置中，所述获取模块501具体用于：对原始图像进行压缩，得到所述压缩后的低动态范围图像。
根据本申请的实施例，在所述装置中，所述分解模块502具体用于：基于Retinex理论对所述低动态范围图像进行分解，得到所述低动态范围图像的细节层和基本层。
根据本申请的实施例，在所述装置中，所述分解模块502进一步用于：对所述低动态范围图像进行保边滤波，将所述保边滤波后得到的图像作为所述低动态范围图像的基本层；并对所述低动态范围图像与基本层图像进行求差，将所述求差后得到的图像作为所述低动态范围图像的细节层。
根据本申请的实施例，在所述装置中，所述第一恢复网络为残差网络，所述第二恢复网络为U-Net网络；所述恢复模块503具体用于：利用所述残差网络对所述细节层进行恢复，并利用所述U-Net网络对所述基本层进行恢复。
根据本申请的实施例，在所述装置中，所述细节层包含所述低动态范围图像的高频分量和压缩伪影，所述基本层包含所述低动态范围图像的低频分量；所述恢复模块503进一步用于：
利用所述残差网络对所述高频分量进行恢复，以便去除所述压缩伪影；并利用所述U-Net网络对所述低频分量进行恢复。
根据本申请的实施例，在所述装置中，所述高频分量包括边缘和轮廓，所述低频分量包括颜色信息和结构信息。
本说明书实施例还提供一种电子设备，包括存储器，处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现上述一种逆色调映射方法。
上述对本说明书特定实施例进行了描述。其它实施例在所附权利要求书的范围内。在一些情况下，在权利要求书中记载的动作或步骤可以按照不同于实施例中的顺序来执行并且仍然可以实现期望的结果。另外，在附图中描绘的过 程不一定要求示出的特定顺序或者连续顺序才能实现期望的结果。在某些实施方式中，多任务处理和并行处理也是可以的或者可能是有利的。
本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于装置、电子设备实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
本说明书实施例提供的装置、电子设备与方法是对应的，因此，装置、电子设备也具有与对应方法类似的有益技术效果，由于上面已经对方法的有益技术效果进行了详细说明，因此，这里不再赘述对应装置、电子设备的有益技术效果。
本说明书是参照根据本说明书实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
还需要说明的是，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、商品或者设备中还存在另外的相同要素。
本说明书可以在由计算机执行的计算机可执行指令的一般上下文中描述，例如程序模块。一般地，程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件、数据结构等等。也可以在分布式计算环境中实践 说明书，在这些分布式计算环境中，由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中，程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
对所公开的实施例的上述说明，使本领域技术人员能够实现或使用本发明。对这些实施例的多种修改对本领域技术人员来说将是显而易见的，本文中所定义的一般原理可以在不脱离本发明的精神和范围的情况下，在其他实施例中实现。因此，本发明将不会被限制于本文所示的这些实施例，而是要符合与本文所公开的原理和新颖特点相一致的最宽的范围。

Claims (16)

  1. 一种逆色调映射方法，所述方法包括：
    获取一个或多个低动态范围图像；
    对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层；
    利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，得到所述恢复后的细节层和基本层；
    利用预定的融合网络对所述恢复后的细节层和基本层进行调整，得到调整后的高动态范围图像。
  2. 如权利要求1所述的方法，所述获取一个或多个低动态范围图像，包括：
    对原始图像进行压缩，得到所述压缩后的低动态范围图像。
  3. 如权利要求2所述的方法，所述对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层，包括：
    基于Retinex理论对所述低动态范围图像进行分解，得到所述低动态范围图像的细节层和基本层。
  4. 如权利要求3所述的方法，所述基于Retinex理论对所述低动态范围图像进行分解，得到所述低动态范围图像的细节层和基本层，具体包括：
    对所述低动态范围图像进行保边滤波，将所述保边滤波后得到的图像作为所述低动态范围图像的基本层；并对所述低动态范围图像与基本层图像进行求差，将所述求差后得到的图像作为所述低动态范围图像的细节层。
  5. 如权利要求1所述的方法，所述第一恢复网络为残差网络，所述第二恢复网络为U-Net网络；所述利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，包括：
    利用所述残差网络对所述细节层进行恢复，并利用所述U-Net网络对所述 基本层进行恢复。
  6. 如权利要求5所述的方法，所述细节层包含所述低动态范围图像的高频分量和压缩伪影，所述基本层包含所述低动态范围图像的低频分量；所述利用所述残差网络对所述细节层进行恢复，并利用所述U-Net网络对所述基本层进行恢复，包括：
    利用所述残差网络对所述高频分量进行恢复，以便去除所述压缩伪影；并利用所述U-Net网络对所述低频分量进行恢复。
  7. 如权利要求6所述的方法，所述高频分量包括边缘和轮廓，所述低频分量包括颜色信息和结构信息。
  8. 如权利要求1所述的方法，所述融合网络采用残差网络。
  9. 一种逆色调映射装置，所述装置包括：
    获取模块，用于获取一个或多个低动态范围图像；
    分解模块，用于对所述低动态范围图像执行分解操作，得到所述低动态范围图像的细节层和基本层；
    恢复模块，用于利用预定的第一恢复网络和第二恢复网络，分别对所述细节层和基本层进行恢复，得到所述恢复后的细节层和基本层；
    调整模块，用于利用预定的融合网络对所述恢复后的细节层和基本层进行调整，得到调整后的高动态范围图像。
  10. 如权利要求9所述的装置，所述获取模块具体用于：
    对原始图像进行压缩，得到所述压缩后的低动态范围图像。
  11. 如权利要求10所述的装置，所述分解模块具体用于：
    基于Retinex理论对所述低动态范围图像进行分解，得到所述低动态范围图像的细节层和基本层。
  12. 如权利要求11所述的装置，所述分解模块进一步用于：
    对所述低动态范围图像进行保边滤波，将所述保边滤波后得到的图像作为所述低动态范围图像的基本层；并对所述低动态范围图像与基本层图像进行求差，将所述求差后得到的图像作为所述低动态范围图像的细节层。
  13. 如权利要求9所述的装置，所述第一恢复网络为残差网络，所述第二恢复网络为U-Net网络；所述恢复模块具体用于：
    利用所述残差网络对所述细节层进行恢复，并利用所述U-Net网络对所述基本层进行恢复。
  14. 如权利要求13所述的装置，所述细节层包含所述低动态范围图像的高频分量和压缩伪影，所述基本层包含所述低动态范围图像的低频分量；所述恢复模块进一步用于：
    利用所述残差网络对所述高频分量进行恢复，以便去除所述压缩伪影；并利用所述U-Net网络对所述低频分量进行恢复。
  15. 如权利要求14所述的装置，所述高频分量包括边缘和轮廓，所述低频分量包括颜色信息和结构信息。
  16. 一种电子设备，包括存储器，处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述程序时实现权利要求1至8中任一项所述的方法。
PCT/CN2019/091874 2019-06-10 2019-06-19 一种逆色调映射方法、装置及电子设备 Ceased WO2020248294A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/975,485 US11783459B2 (en) 2019-06-10 2019-06-19 Method and device of inverse tone mapping and electronic device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910499995.3 2019-06-10
CN201910499995.3A CN110223256A (zh) 2019-06-10 2019-06-10 一种逆色调映射方法、装置及电子设备

Publications (1)

Publication Number Publication Date
WO2020248294A1 true WO2020248294A1 (zh) 2020-12-17

Family

ID=67816251

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/091874 Ceased WO2020248294A1 (zh) 2019-06-10 2019-06-19 一种逆色调映射方法、装置及电子设备

Country Status (3)

Country Link
US (1) US11783459B2 (zh)
CN (1) CN110223256A (zh)
WO (1) WO2020248294A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114581336A (zh) * 2022-03-21 2022-06-03 北京有竹居网络技术有限公司 图像修复方法、装置、设备、介质及产品
CN119067967A (zh) * 2024-11-04 2024-12-03 四川公路桥梁建设集团有限公司 一种基于人工智能的桥梁的挠度智能化精确测量方法

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3944615A1 (en) * 2015-09-02 2022-01-26 Faurecia Irystec Inc. System and method for real-time tone-mapping
CN110796595B (zh) * 2019-10-31 2022-03-01 北京大学深圳研究生院 一种色调映射方法、装置及电子设备
US11893482B2 (en) * 2019-11-14 2024-02-06 Microsoft Technology Licensing, Llc Image restoration for through-display imaging
CN113628123B (zh) * 2020-05-09 2026-01-13 深圳市中兴微电子技术有限公司 图像恢复模型的训练方法及装置、电子设备、可读介质
CN111683269B (zh) * 2020-06-12 2021-08-17 腾讯科技（深圳）有限公司 视频处理方法、装置、计算机设备和存储介质
CN112991209B (zh) * 2021-03-12 2024-01-12 北京百度网讯科技有限公司 图像处理方法、装置、电子设备及存储介质
CN117716385A (zh) * 2021-07-29 2024-03-15 杜比实验室特许公司 用于图像的动态范围转换和显示管理的神经网络
CN114429430B (zh) * 2022-01-30 2024-11-15 京东方科技集团股份有限公司 图像处理方法、电子设备和非瞬态计算机可读介质
CN117830183B (zh) * 2024-03-05 2024-05-14 俐玛精密测量技术(苏州)有限公司 一种针对ct图像的色调映射方法和相关装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105850114A (zh) * 2013-12-27 2016-08-10 汤姆逊许可公司 用于图像的逆色调映射的方法
CN107545547A (zh) * 2016-06-28 2018-01-05 硅工厂股份有限公司 逆色调映射方法
CN107949864A (zh) * 2015-08-25 2018-04-20 汤姆逊许可公司 基于亮度区域的逆色调映射
CN109785263A (zh) * 2019-01-14 2019-05-21 北京大学深圳研究生院 一种基于Retinex的逆色调映射图像转换方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8121401B2 (en) * 2006-01-24 2012-02-21 Sharp Labortories of America, Inc. Method for reducing enhancement of artifacts and noise in image color enhancement
KR20120114899A (ko) * 2011-04-08 2012-10-17 삼성전자주식회사 영상 처리 방법 및 영상 처리 장치
EP3944615A1 (en) * 2015-09-02 2022-01-26 Faurecia Irystec Inc. System and method for real-time tone-mapping
EP3418972A1 (en) * 2017-06-23 2018-12-26 Thomson Licensing Method for tone adapting an image to a target peak luminance lt of a target display device
US20210166360A1 (en) * 2017-12-06 2021-06-03 Korea Advanced Institute Of Science And Technology Method and apparatus for inverse tone mapping

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105850114A (zh) * 2013-12-27 2016-08-10 汤姆逊许可公司 用于图像的逆色调映射的方法
CN107949864A (zh) * 2015-08-25 2018-04-20 汤姆逊许可公司 基于亮度区域的逆色调映射
CN107545547A (zh) * 2016-06-28 2018-01-05 硅工厂股份有限公司 逆色调映射方法
CN109785263A (zh) * 2019-01-14 2019-05-21 北京大学深圳研究生院 一种基于Retinex的逆色调映射图像转换方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114581336A (zh) * 2022-03-21 2022-06-03 北京有竹居网络技术有限公司 图像修复方法、装置、设备、介质及产品
CN119067967A (zh) * 2024-11-04 2024-12-03 四川公路桥梁建设集团有限公司 一种基于人工智能的桥梁的挠度智能化精确测量方法

Also Published As

Publication number Publication date
US20230025557A1 (en) 2023-01-26
US11783459B2 (en) 2023-10-10
CN110223256A (zh) 2019-09-10

Similar Documents

Publication Publication Date Title
WO2020248294A1 (zh) 一种逆色调映射方法、装置及电子设备
CN110796595B (zh) 一种色调映射方法、装置及电子设备
CN105825472A (zh) 一种快速的基于多尺度高斯滤波器的色调映射系统及方法
CN105850114A (zh) 用于图像的逆色调映射的方法
CN103400342A (zh) 基于混合色阶映射及压缩系数的高动态范围图像重构方法
WO2022166298A1 (zh) 一种图像处理方法、装置、电子设备及可读存储介质
CN110428382B (zh) 一种用于移动终端的高效视频增强方法、装置和存储介质
CN109785263B (zh) 一种基于Retinex的逆色调映射图像转换方法
CN105243641A (zh) 一种基于双树复小波变换的低光照图像增强方法
CN112927160A (zh) 一种基于深度Retinex的单张低光图像增强方法
CN114998138A (zh) 一种基于注意力机制的高动态范围图像去伪影方法
CN110572652A (zh) 一种静态图像的处理方法及装置
CN112102166B (zh) 一种联合超分辨率、色域扩展和逆色调映射方法及设备
CN106709888B (zh) 一种基于人眼视觉模型的高动态范围图像产生方法
CN111091515B (zh) 图像复原方法及装置、计算机可读存储介质
JP7627389B2 (ja) Ｓｄｒ－ｈｄｒ局所的再構成のためのノイズ除去
JP2020028111A (ja) 画像処理方法、画像処理装置、画像処理システム、およびプログラム
CN105894456B (zh) 一种基于正规化分层的高动态范围图像阶调映射方法
CN109102463B (zh) 一种超分辨率图像重建方法及装置
WO2024055458A1 (zh) 图像降噪处理方法、装置、设备、存储介质和程序产品
CN113724144B (zh) 一种图像处理方法及终端设备上的图像信号处理器
CN105516674A (zh) 一种具有hdr显示功能的监视设备
CN110175959B (zh) 一种台风云图增强方法
JP7761780B2 (ja) Ｒａｗ画像データ再構成システムおよび方法
US10438328B1 (en) Chroma blurring reduction in video and images

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19932516

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19932516

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19932516

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 08.08.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19932516

Country of ref document: EP

Kind code of ref document: A1