WO2020118977A1 - 图像色温校正方法及装置 - Google Patents

图像色温校正方法及装置 Download PDF

Info

Publication number
WO2020118977A1
WO2020118977A1 PCT/CN2019/081324 CN2019081324W WO2020118977A1 WO 2020118977 A1 WO2020118977 A1 WO 2020118977A1 CN 2019081324 W CN2019081324 W CN 2019081324W WO 2020118977 A1 WO2020118977 A1 WO 2020118977A1
Authority
WO
WIPO (PCT)
Prior art keywords
color temperature
image
color
skin
processor
Prior art date
Application number
PCT/CN2019/081324
Other languages
English (en)
French (fr)
Inventor
濮怡莹
饶洋
许神贤
金羽锋
Original Assignee
深圳市华星光电半导体显示技术有限公司
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 深圳市华星光电半导体显示技术有限公司 filed Critical 深圳市华星光电半导体显示技术有限公司
Publication of WO2020118977A1 publication Critical patent/WO2020118977A1/zh

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N9/00Details of colour television systems
    • H04N9/64Circuits for processing colour signals
    • H04N9/73Colour balance circuits, e.g. white balance circuits or colour temperature control

Definitions

  • the invention relates to an image processing technology, in particular to an image color temperature correction method and device.
  • each pixel on the image sensor depends not only on the scene itself, but also on the light source. That is, the same scene will show different image colors under different light source temperatures.
  • the automatic mode of white balance of most cameras or monitors compensates the image by estimating the color temperature of the light source of the image to restore the original color of the scene.
  • the invention provides an image color temperature correction method and device to solve the problem of misjudgment caused by the unrecognized image content of the color temperature statistical algorithm in the prior art.
  • An aspect of the present invention provides an image color temperature correction method.
  • the method can be executed by a processor.
  • the processor can be coupled to a memory.
  • the memory can include at least one program instruction.
  • the program instruction can be executed by the Executed by a processor, the method may include the steps of: calculating, by the processor, an original color temperature data of an image; identifying, by the processor, at least one memory color object in the image, and calculating the at least one A memory color object corresponding to at least one object color temperature data; the processor calculating a corrected color temperature data based on the original color temperature data, the at least one object color temperature data and at least one weight data; and based on the processor
  • the corrected color temperature data adjusts a color temperature at which the image is output; wherein the calculation method of the corrected color temperature data is as follows:
  • CT fnl is the corrected color temperature data
  • CT 0 is the original color temperature data
  • ⁇ 0 is an original weight data
  • CT i is the object color temperature data
  • ⁇ i is the weight data
  • the ⁇ 0 and The sum of all ⁇ i is 1.
  • the memory color object may be selected from a group consisting of a skin color recognition object, a green plant recognition object, or a sun height recognition object.
  • a scene distinction and object segmentation method is applied to the image to generate a semantic label map containing multiple masks, and the semantic A mask about a portrait in the label diagram, a portrait component is extracted from the image using the mask, a skin color part is extracted from the portrait component using a common skin color range, and the skin color part is generated according to a color temperature statistical result
  • An initial skin tone color temperature, the initial skin tone color temperature is adjusted to a corrected skin tone color temperature as the target color temperature data, and the calculation method of the corrected skin tone color temperature is as follows:
  • CT corrskin CT m +CT skin -CT averskin
  • CT corrskin is the corrected skin color temperature
  • CT m is a neutral color temperature constant
  • CT skin is the initial skin color temperature
  • CT averskin is a skin color average color temperature constant.
  • a scene distinction and object segmentation method is applied to the image to generate a semantic label map containing multiple masks, extracted in the A mask for grass and trees in the semantic label diagram, a green plant component is extracted from the image with the mask, and the green plant component generates a green plant color temperature as the target color temperature data according to a color temperature statistical result.
  • a sun brightness and a sun height angle are extracted from the image by a neural network algorithm for sun height recognition.
  • a value is 1 or 0, a value range of the solar altitude angle is 0 degrees to 90 degrees, the solar brightness and the solar altitude angle are generated according to a solar altitude color temperature table look-up table to generate a solar altitude color temperature as The object color temperature data.
  • the device includes a processor and a memory.
  • the memory is coupled to the processor and includes at least one program instruction.
  • the program instruction can be processed by the processor.
  • the device also includes:
  • a statistical color temperature module configured to cause the processor to calculate an original color temperature data of an image
  • a target recognition module configured to cause the processor to recognize at least one memory color object in the image and calculate at least one object color temperature data corresponding to the at least one memory color object
  • a color temperature correction module configured to cause the processor to calculate a corrected color temperature data based on the original color temperature data, the at least one object color temperature data, and at least one weight data, and adjust the image according to the corrected color temperature data The output color temperature;
  • the calculation method of the corrected color temperature data is as follows:
  • CT fnl is the corrected color temperature data
  • CT 0 is the original color temperature data
  • ⁇ 0 is an original weight data
  • CT i is the object color temperature data
  • ⁇ i is the weight data
  • the ⁇ 0 and The sum of all ⁇ i is 1;
  • the target recognition module includes a skin color recognition unit configured to cause the processor to recognize a skin color recognition object in the image;
  • the target recognition module includes a green plant recognition unit configured to cause the processor to recognize a green plant recognition object in the image;
  • the target recognition module includes a sun height recognition unit configured to cause the processor to recognize a sun height recognition object in the image.
  • the skin color recognition unit is configured to cause the processor to apply a scene distinction and object segmentation method to the image to generate a semantic label image containing multiple masks, and extract the semantic label
  • a portrait component is extracted from the image using the mask
  • a skin color portion is extracted from the portrait component using a common skin color range
  • the skin color portion is generated according to a color temperature statistical result
  • the initial skin tone color temperature, the initial skin tone color temperature is adjusted to a corrected skin tone color temperature as the target color temperature data, and the calculation method of the corrected skin tone color temperature is as follows:
  • CT corrskin CT m +CT skin -CT averskin
  • CT corrskin is the corrected skin color temperature
  • CT m is a neutral color temperature constant
  • CT skin is the initial skin color temperature
  • CT averskin is a skin color average color temperature constant.
  • the green plant recognition unit is configured to cause the processor to apply a scene distinction and object segmentation method to the image to generate a semantic label map containing multiple masks, extracting the semantic A mask for grass and trees in the label diagram, a green plant component is extracted from the image with the mask, and a green plant color temperature is generated from the green plant component according to a color temperature statistical result as the target color temperature data.
  • the solar altitude recognition unit is configured to cause the processor to extract a solar brightness and a solar altitude angle from the image using a neural network algorithm for solar altitude recognition, a The value is 1 or 0, and a range of the solar altitude angle is 0 degrees to 90 degrees.
  • the solar brightness and the solar altitude angle are generated according to a solar altitude color temperature table lookup table to generate a solar altitude color temperature as a Describe the target color temperature data.
  • the device includes a processor and a memory.
  • the memory is coupled to the processor and includes at least one program instruction.
  • the program instruction can be processed by the processor.
  • the device further includes: a statistical color temperature module configured to cause the processor to calculate an original color temperature data of an image; and a target recognition module configured to cause the processor to evaluate the At least one memory color object to identify and calculate at least one object color temperature data corresponding to the at least one memory color object; and a color temperature correction module configured to cause the processor to base on the original color temperature data, the At least one object color temperature data and at least one weight data calculate a corrected color temperature data, and adjust a color temperature of the image to be output according to the corrected color temperature data; wherein the calculation method of the corrected color temperature data is as follows:
  • CT fnl is the corrected color temperature data
  • CT 0 is the original color temperature data
  • ⁇ 0 is an original weight data
  • CT i is the object color temperature data
  • ⁇ i is the weight data
  • the ⁇ 0 and The sum of all ⁇ i is 1.
  • the target recognition module includes a skin color recognition unit configured to cause the processor to recognize a skin color recognition object in the image.
  • the skin color recognition unit is configured to cause the processor to apply a scene distinction and object segmentation method to the image to generate a semantic label image containing multiple masks, and extract the semantic label
  • a portrait component is extracted from the image using the mask
  • a skin color portion is extracted from the portrait component using a common skin color range
  • the skin color portion is generated according to a color temperature statistical result
  • the initial skin tone color temperature, the initial skin tone color temperature is adjusted to a corrected skin tone color temperature as the target color temperature data, and the calculation method of the corrected skin tone color temperature is as follows:
  • CT corrskin CT m +CT skin -CT averskin
  • CT corrskin is the corrected skin color temperature
  • CT m is a neutral color temperature constant
  • CT skin is the initial skin color temperature
  • CT averskin is a skin color average color temperature constant.
  • the target identification module includes a green plant identification unit configured to cause the processor to identify a green plant identification object in the image.
  • the green plant recognition unit is configured to cause the processor to apply a scene distinction and object segmentation method to the image to generate a semantic label map containing multiple masks, extracting the semantic A mask for grass and trees in the label diagram, a green plant component is extracted from the image with the mask, and a green plant color temperature is generated from the green plant component according to a color temperature statistical result as the target color temperature data.
  • the target recognition module includes a solar altitude recognition unit configured to cause the processor to recognize a solar altitude recognition object in the image.
  • the solar altitude recognition unit is configured to cause the processor to extract a solar brightness and a solar altitude angle from the image using a neural network algorithm for solar altitude recognition, a The value is 1 or 0, and a range of the solar altitude angle is 0 degrees to 90 degrees.
  • the solar brightness and the solar altitude angle are generated according to a solar altitude color temperature table lookup table to generate a solar altitude color temperature as a Describe the target color temperature data.
  • the image color temperature correction method and device of the present invention preliminary estimate the image color temperature by using a statistical method, then use the target recognition method to identify special objects (such as different memory color objects) and calculate the color temperature, and finally estimate the color temperature Correction is performed to obtain the corrected image color temperature, which corresponds to the memory color object contained in the image, and the beneficial effects of improving the accuracy of image color temperature estimation and making the color temperature correction result closer to human observation results can be obtained.
  • FIG. 1 is a schematic diagram of an image color temperature correction device according to an embodiment of the invention.
  • FIG. 2 is a schematic diagram of a color temperature partition according to an embodiment of the invention.
  • FIG. 3 is a schematic diagram of semantic tags according to an embodiment of the present invention.
  • an image color temperature correction device may include a processor and a memory, the memory is coupled to the processor and includes at least one program instruction, The program instructions can be executed by the processor.
  • the image color temperature correction device may further include a statistical color temperature module 1, a target recognition module 2, and a color temperature correction module 3. The following is an example, but not limited to this.
  • the statistical color temperature module 1 is configured to cause the processor to calculate an original color temperature data of an image.
  • the image color temperature correction device may further include an imaging module (for example: a camera, etc.) and a communication module (for example: various wireless communication transceiver modules, etc.), the imaging module and the communication
  • the module may be coupled to the processor, and the image capturing module and the communication module may be configured to cause the processor to obtain a content data of the image, for example, the raw data of the image Wait.
  • the target recognition module 2 may be configured to cause the processor to recognize at least one memory color object in the image and calculate the corresponding value of the at least one memory color object At least one object color temperature data.
  • the target recognition module 2 may include a skin color recognition unit 21, which may be configured to cause the processor to recognize a skin color recognition object in the image, for example:
  • the skin color recognition object may be various portrait skin colors.
  • the skin color recognition unit 21 may be configured to cause the processor to apply a scene distinction and object segmentation method to the image to generate a semantic label map containing multiple masks ), extract a mask about a portrait in the semantic label diagram, extract a portrait component from the mask using the mask, extract a skin color part from the portrait component using a common skin color range, and extract the skin color
  • An initial skin tone color temperature is generated based in part on a color temperature statistical result, and the initial skin tone color temperature is adjusted to a corrected skin tone color temperature as the target color temperature data.
  • the calculation method of the corrected skin tone color temperature is as follows:
  • CT corrskin CT m +CT skin -CT averskin
  • CT corrskin is the corrected skin color temperature
  • CT m is a neutral color temperature constant
  • CT skin is the initial skin color temperature
  • CT averskin is a skin color average color temperature constant.
  • the target recognition module 2 may include a green plant recognition unit 22, which may be configured to cause the processor to recognize a green plant recognition object in the image
  • the green plant identification object may be green plants such as grass or trees.
  • the green plant recognition unit 22 may be configured to cause the processor to apply a scene distinction and object segmentation method to the image to generate a semantic label map containing multiple masks, extracting the semantics A mask for grass and trees in the label diagram, a green plant component is extracted from the image with the mask, and a green plant color temperature is generated from the green plant component according to a color temperature statistical result as the target color temperature data.
  • the target recognition module 2 may include a solar altitude recognition unit 23, and the solar altitude recognition unit 23 may be configured to cause the processor to recognize a solar altitude recognition object in the image
  • the object for identifying the height of the sun may be an image feature possessed by sunlight such as sunrise, midday, or sunset.
  • the solar altitude recognition unit 23 may be configured to cause the processor to extract a solar brightness and a solar altitude angle from the image using a neural network algorithm for solar altitude recognition, a The value is 1 or 0, and a range of the solar altitude angle is 0 degrees to 90 degrees.
  • the solar brightness and the solar altitude angle are generated according to a solar altitude color temperature table lookup table to generate a solar altitude color temperature as a Describe the target color temperature data.
  • the color temperature correction module 3 is configured to cause the processor to calculate a corrected color temperature data based on the original color temperature data, the at least one object color temperature data, and at least one weight data, and according to the The corrected color temperature data adjusts a color temperature at which the image is output.
  • the calculation method of the corrected color temperature data is as follows:
  • CT fnl is the corrected color temperature data
  • CT 0 is the original color temperature data
  • ⁇ 0 is an original weight data
  • CT i is the object color temperature data
  • ⁇ i is the weight data
  • the ⁇ 0 and The sum of all ⁇ i is 1.
  • the ⁇ 0 and all ⁇ i can be extracted according to a large amount of data collected in advance or set manually.
  • the image color temperature correction device may also be configured as a part of a device with image data processing functions, such as a functional module in a notebook computer, tablet computer, or smartphone, to assist in processing related images before being output Color temperature correction function.
  • image data processing functions such as a functional module in a notebook computer, tablet computer, or smartphone
  • an image color temperature correction method can be executed by a processor.
  • the processor can be coupled to a memory.
  • the memory can include at least one program instruction.
  • the program instruction can Performed by the processor, the method may include the steps of: calculating, by the processor, an original color temperature data of an image; identifying, by the processor, at least one memory color object in the image, and calculating At least one object color temperature data corresponding to the at least one memory color object; a corrected color temperature data calculated by the processor based on the original color temperature data, the at least one object color temperature data and at least one weight data; and by the The processor adjusts a color temperature at which the image is output according to the corrected color temperature data; wherein the calculation method of the corrected color temperature data is as follows:
  • CT fnl is the corrected color temperature data
  • CT 0 is the original color temperature data
  • ⁇ 0 is an original weight data
  • CT i is the object color temperature data
  • ⁇ i is the weight data
  • the ⁇ 0 and The sum of all ⁇ i is 1.
  • the following examples illustrate the implementation of extracting the original color temperature data, but not limited thereto.
  • the original color temperature data can be extracted from the image according to a gray world algorithm (Gray World Algorithm), and the gray world algorithm can refer to the comparison of algorithms for calculating color constancy—Part 1: Method of synthesizing data And the experimental report, as shown in the following formula:
  • ⁇ 1 , ⁇ 2 , and ⁇ 3 are the three-channel digital values of red (r), green (g), and blue (b); n is the total number of pixels.
  • a ⁇ -CT table obtained in advance from a test experiment is queried to obtain a CT value as the original color temperature data.
  • the original color temperature data can also be obtained by the following method steps: calculating the r, g, and b values of the image pixels, and converting the r, g, and b values into R, G, and B by gamma transformation Optical value (according to CIE conversion formula), convert R, G, B optical values into X, Y, Z tristimulus values (according to CIE conversion formula) through the display's TM (transformation) matrix, convert X, Y, Z tristimulus The value is converted into the x and y values of the CIE1931 space (according to the CIE conversion formula).
  • the color temperature partition of a color point is determined based on the x and y values (as shown in Figure 2 It can be understood by those skilled in the art that the color temperature map can also be displayed in color), and the calculation is performed according to different color temperature partitions. Examples are as follows:
  • CT -437*n ⁇ 3+3601*n ⁇ 2-6861*n+5514.31.
  • n (x PA -0.332) / (y PA -0.1858), x PA, y PA is the color point P A x in FIG. 2, y values.
  • the memory color object may be selected from a group consisting of a skin color recognition object, a green plant recognition object, or a sun height recognition object.
  • a scene distinction and object segmentation method is applied to the image to generate a semantic label map containing multiple masks (such as various deep neural network-based The algorithm of automatic dimming of the network is obtained), extract the mask about a portrait in the semantic label diagram, (such as M person , such as the cross-sectional area of "person" in FIG.
  • the calculation method of the corrected skin color temperature is as follows:
  • CT corrskin CT m +CT skin -CT averskin
  • CT corrskin is the corrected skin color temperature
  • CT m is a neutral color temperature constant (generally 4000K to 5500K)
  • CT skin is the initial skin color temperature
  • CT averskin is a skin color average color temperature constant Picture calculation).
  • the reason for the skin color recognition is that the skin color belongs to the memory color, but it is in the warm color range.
  • the skin color is extracted separately for later correction of the statistical color temperature .
  • green plants are memory colors relative to other outdoor scenes such as flowers.
  • the green plants are extracted separately to increase the weight of green plants. To correct the statistical color temperature.
  • a sun brightness is extracted from the image using a neural network algorithm for sun height recognition (such as various algorithms for learning high dynamic range from outdoor panoramas) (such as I sun ) and a solar altitude angle (such as y sun ), a value of the solar brightness is 1 or 0, and a value range of the solar height angle is 0 degrees to 90 degrees (°, degree; 0 ° is close to sunrise or sunset, 90° is close to noon), the solar brightness and the solar altitude angle and a solar altitude color temperature table (as shown in Table 2 below, taking the equatorial area data as an example, the data in other areas need to be based on The solar altitude angle is converted) and the comparison generates a solar altitude color temperature (such as CT sun ) as the target color temperature data.
  • a neural network algorithm for sun height recognition such as various algorithms for learning high dynamic range from outdoor panoramas
  • a solar altitude angle such as y sun
  • a value of the solar brightness is 1 or 0
  • a value range of the solar height angle
  • the image color temperature correction method and device embodiment of the present invention uses a statistical method to initially estimate the image color temperature, and then uses a target recognition method to identify special objects (such as different memory color objects) and calculate the color temperature, and finally corrects the estimated color temperature to obtain a correction After the image color temperature.
  • the existing image color temperature statistical algorithm is used to process the image color temperature, in a sunset image with a large area of blue sky, since blue accounts for most of the picture, the statistical results are mainly based on the image color temperature is high (partial Cold), but in actual scenes, people feel that the sunset light source is low (warm); in addition, the portrait image under the cold light source, because the skin color is generally warm, if the skin color accounts for most of the image area, the statistics As a result, the color temperature of the image is low (warm), but in actual scenes, the light source that people feel is high (cold); and, in the image with a large area of monochromatic red flowers, because the red is warm, the statistical results The color temperature of the image is low (warm), but people can judge the scene light source to be neutral based on the memory color of the green leaves.
  • the color temperature correction result of the present invention corresponds to the memory color objects (such as sky, skin color, green plants, etc.) contained in the image, which can improve the accuracy of image color temperature estimation Bring the color temperature correction result closer to the human observation result and other beneficial effects.
  • the memory color objects such as sky, skin color, green plants, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

一种图像色温校正方法及装置,所述装置包括一处理器及一内存,所述装置还包括:一统计色温模块,被配置成致使一处理器计算一图像的一原始色温数据;一目标识别模块,被配置成致使所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据;及一色温校正模块,被配置成致使所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据,及依据所述校正色温数据调整所述图像被输出的一色温。

Description

图像色温校正方法及装置 技术领域
本发明是有关于一种图像处理技术,特别是有关于一种图像色温校正方法及装置。
背景技术
当采用数码相机捕获一个场景的图像时,在图像传感器上每个像素的响应除了取决于场景本身,也会受到光源的影响。即同一个场景,在不同的光源温度下,会呈现不同的图像色彩。
举例来说,当一个白色物体在低色温下被照亮时会呈现红色,而在高色温下会呈现蓝色。因此,大部分相机或显示器的白平衡的自动模式通过估计图像的光源色温对图像进行补偿,以还原场景的原本色彩。
现有的图像色温统计算法大部分是以算法对图像的色彩分布进行假设,从而进一步统计和估计照明光源色温。但是这种单纯基于统计的方法强烈地依赖于算法对色彩分布的假设,当图像不满足算法的假设时,计算出的色温结果就会产生偏差。同时,这类统计的方法对图像内容未进行识别,容易导致色温误判。
因此,现有技术存在缺陷,急需改进。
技术问题
本发明提供一种图像色温校正方法及装置,以解决现有技术所存在的色温统计算法未识别图像内容导致误判的问题。
技术解决方案
本发明的一方面提供一种图像色温校正方法,所述方法可由一处理器执行,所述处理器可耦接一内存,所述内存可包括至少一程序指令,所述程序指令能由所述处理器执行,所述方法可包括步骤:由所述处理器计算一图像的一原始色温数据;由所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据;由所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据;及由所述处理器依据所述校正色温数据调整所述图像被输出的一色温;其中所述校正色温数据的计算方式,如下式所示:
Figure PCTCN2019081324-appb-000001
其中,CT fnl为所述校正色温数据,CT 0为所述原始色温数据,α 0为一原始权重数据,CT i为所述对象色温数据,α i为所述权重数据,所述α 0与所有α i的总和为1。
在一些实施例中,所述记忆色对象可选自由一肤色识别对象、一绿植识别对象或一太阳高度识别对象组成的一群组。
在一些实施例中,当所述记忆色对象为所述肤色识别对象时,对所述图像采用一场景区别及物体分割方法以产 生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于一人像的掩膜,以所述掩膜对所述图像提取一人像分量,采用一常用肤色范围对所述人像分量提取一肤色部分,将所述肤色部分依据一色温统计结果产生一初始肤色色温,将所述初始肤色色温调整为一校正肤色色温作为所述对象色温数据,所述校正肤色色温的计算方式如下式所示:
CT corrskin=CT m+CT skin-CT averskin
其中,CT corrskin为所述校正肤色色温,CT m为一中性色温常量,CT skin为所述初始肤色色温,CT averskin为一肤色平均色温常量。
在一些实施例中,当所述记忆色对象为所述绿植识别对象时,对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于草及树的掩膜,以所述掩膜对所述图像提取一绿植分量,将所述绿植分量依据一色温统计结果产生一绿植色温作为所述对象色温数据。
在一些实施例中,当所述记忆色对象为所述太阳高度识别对象时,以一太阳高度识别的神经网路算法对所述图像提取一太阳亮度及一太阳高度角,所述太阳亮度的一取值为1或0,所述太阳高度角的一取值范围为0度至90度,将所述太阳亮度及所述太阳高度角依据一太阳高度色 温表查表产生一太阳高度色温作为所述对象色温数据。
本发明的另一方面提供一种图像色温校正装置,所述装置包括一处理器及一内存,所述内存耦接所述处理器且包括至少一程序指令,所述程序指令能由所述处理器执行,所述装置还包括:
一统计色温模块,被配置成致使所述处理器计算一图像的一原始色温数据;
一目标识别模块,被配置成致使所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据;及
一色温校正模块,被配置成致使所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据,及依据所述校正色温数据调整所述图像被输出的一色温;
其中所述校正色温数据的计算方式,如下式所示:
Figure PCTCN2019081324-appb-000002
其中,CT fnl为所述校正色温数据,CT 0为所述原始色温数据,α 0为一原始权重数据,CT i为所述对象色温数据,α i为所述权重数据,所述α 0与所有α i的总和为1;
所述目标识别模块包括一肤色识别单元,所述肤色识别单元被配置成致使所述处理器对于所述图像中的一肤 色识别对象进行识别;
所述目标识别模块包括一绿植识别单元,所述绿植识别单元被配置成致使所述处理器对于所述图像中的一绿植识别对象进行识别;及
所述目标识别模块包括一太阳高度识别单元,所述太阳高度识别单元被配置成致使所述处理器对于所述图像中的一太阳高度识别对象进行识别。
在一些实施例中,所述肤色识别单元被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于一人像的掩膜,以所述掩膜对所述图像提取一人像分量,采用一常用肤色范围对所述人像分量提取一肤色部分,将所述肤色部分依据一色温统计结果产生一初始肤色色温,将所述初始肤色色温调整为一校正肤色色温作为所述对象色温数据,所述校正肤色色温的计算方式如下式所示:
CT corrskin=CT m+CT skin-CT averskin
其中,CT corrskin为所述校正肤色色温,CT m为一中性色温常量,CT skin为所述初始肤色色温,CT averskin为一肤色平均色温常量。
在一些实施例中,所述绿植识别单元被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以 产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于草及树的掩膜,以所述掩膜对所述图像提取一绿植分量,将所述绿植分量依据一色温统计结果产生一绿植色温作为所述对象色温数据。
在一些实施例中,所述太阳高度识别单元被配置成致使所述处理器以一太阳高度识别的神经网路算法对所述图像提取一太阳亮度及一太阳高度角,所述太阳亮度的一取值为1或0,所述太阳高度角的一取值范围为0度至90度,将所述太阳亮度及所述太阳高度角依据一太阳高度色温表查表产生一太阳高度色温作为所述对象色温数据。
本发明的另一方面提供一种图像色温校正装置,所述装置包括一处理器及一内存,所述内存耦接所述处理器且包括至少一程序指令,所述程序指令能由所述处理器执行,所述装置还包括:一统计色温模块,被配置成致使所述处理器计算一图像的一原始色温数据;一目标识别模块,被配置成致使所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据;及一色温校正模块,被配置成致使所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据,及依据所述校正色温数据调整所述图像被输出的一色温;其中所述校正色温数据的计算方式,如下式所示:
Figure PCTCN2019081324-appb-000003
其中,CT fnl为所述校正色温数据,CT 0为所述原始色温数据,α 0为一原始权重数据,CT i为所述对象色温数据,α i为所述权重数据,所述α 0与所有α i的总和为1。
在一些实施例中,所述目标识别模块包括一肤色识别单元,所述肤色识别单元被配置成致使所述处理器对于所述图像中的一肤色识别对象进行识别。
在一些实施例中,所述肤色识别单元被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于一人像的掩膜,以所述掩膜对所述图像提取一人像分量,采用一常用肤色范围对所述人像分量提取一肤色部分,将所述肤色部分依据一色温统计结果产生一初始肤色色温,将所述初始肤色色温调整为一校正肤色色温作为所述对象色温数据,所述校正肤色色温的计算方式如下式所示:
CT corrskin=CT m+CT skin-CT averskin
其中,CT corrskin为所述校正肤色色温,CT m为一中性色温常量,CT skin为所述初始肤色色温,CT averskin为一肤色平均色温常量。
在一些实施例中,所述目标识别模块包括一绿植识别 单元,所述绿植识别单元被配置成致使所述处理器对于所述图像中的一绿植识别对象进行识别。
在一些实施例中,所述绿植识别单元被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于草及树的掩膜,以所述掩膜对所述图像提取一绿植分量,将所述绿植分量依据一色温统计结果产生一绿植色温作为所述对象色温数据。
在一些实施例中,所述目标识别模块包括一太阳高度识别单元,所述太阳高度识别单元被配置成致使所述处理器对于所述图像中的一太阳高度识别对象进行识别。
在一些实施例中,所述太阳高度识别单元被配置成致使所述处理器以一太阳高度识别的神经网路算法对所述图像提取一太阳亮度及一太阳高度角,所述太阳亮度的一取值为1或0,所述太阳高度角的一取值范围为0度至90度,将所述太阳亮度及所述太阳高度角依据一太阳高度色温表查表产生一太阳高度色温作为所述对象色温数据。
有益效果
与现有技术相比较,本发明的图像色温校正方法及装置通过采用统计方法对图像色温进行初步估计,再采用目标识别法识别特殊物体(如不同记忆色对象)并计算色温,最后对估计色温进行校正,得到校正后的影像色温,所述校正后的影像色温相应于图像所含的记忆色对象,可以获 得提高影像色温估计的准确率,使色温校正结果更加接近人为观察结果等有益效果。
附图说明
图1是本发明的一实施例的图像色温校正装置的示意图。
图2是本发明的一实施例的色温分区的示意图。
图3是本发明的一实施例的语义标签的示意图。
本发明的最佳实施方式
以下各实施例的说明是参考附加的图式,用以例示本发明可用以实施的特定实施例。再者,本发明所提到的方向用语,例如上、下、顶、底、前、后、左、右、内、外、侧面、周围、中央、水平、横向、垂直、纵向、轴向、径向、最上层或最下层等,仅是参考附加图式的方向。因此,使用的方向用语是用以说明及理解本发明,而非用以限制本发明。
请参照图1所示,本发明的一实施例的图像色温校正装置,可包括一处理器(processor)及一内存(memory),所述内存耦接所述处理器且包括至少一程序指令,所述程序指令能由所述处理器执行。所述图像色温校正装置还可包括一统计色温模块1、一目标识别模块2及一色温校正模块3。举例说明如下,但不以此为限。
请再参照图1所示,所述统计色温模块1被配置成致使所述处理器计算一图像的一原始色温数据。
在一些实施例中,所述图像色温校正装置还可包括一取像模块(例如:摄像头等)及一通信模块(例如:各种无线通信收发模块等),所述取像模块及所述通信模块可耦接所述处理器,并且所述取像模块及所述通信模块可被配置成致使所述处理器取得所述图像的一内容数据,例如:所述图像的原始数据(raw data)等。
请再参照图1所示,所述目标识别模块2可被配置成致使所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据。
在一些实施例中,所述目标识别模块2可包括一肤色识别单元21,所述肤色识别单元21可被配置成致使所述处理器对于所述图像中的一肤色识别对象进行识别,例如:所述肤色识别对象可为各种人像肤色等。
举例来说,所述肤色识别单元21可被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜(mask)的一语义标签图(semantic label map),提取在所述语义标签图中关于一人像的掩膜,以所述掩膜对所述图像提取一人像分量,采用一常用肤色范围对所述人像分量提取一肤色部分,将所述肤色部分依据一色温统计结果产生一初始肤色色温,将所述初始肤色色温调整为一校正肤色色温作为所述对象色温数据,所述校正 肤色色温的计算方式如下式所示:
CT corrskin=CT m+CT skin-CT averskin
其中,CT corrskin为所述校正肤色色温,CT m为一中性色温常量,CT skin为所述初始肤色色温,CT averskin为一肤色平均色温常量。
在一些实施例中,所述目标识别模块2可包括一绿植识别单元22,所述绿植识别单元22可被配置成致使所述处理器对于所述图像中的一绿植识别对象进行识别,例如:绿植识别对象可为草或树等绿色植物。
举例来说,所述绿植识别单元22可被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于草及树的掩膜,以所述掩膜对所述图像提取一绿植分量,将所述绿植分量依据一色温统计结果产生一绿植色温作为所述对象色温数据。
在一些实施例中,所述目标识别模块2可包括一太阳高度识别单元23,所述太阳高度识别单元23可被配置成致使所述处理器对于所述图像中的一太阳高度识别对象进行识别,例如:所述太阳高度识别对象可为日出、日正当中或夕阳等阳光具备的图像特征。
举例来说,所述太阳高度识别单元23可被配置成致使所述处理器以一太阳高度识别的神经网路算法对所述 图像提取一太阳亮度及一太阳高度角,所述太阳亮度的一取值为1或0,所述太阳高度角的一取值范围为0度至90度,将所述太阳亮度及所述太阳高度角依据一太阳高度色温表查表产生一太阳高度色温作为所述对象色温数据。
请再参照图1所示,所述色温校正模块3被配置成致使所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据,及依据所述校正色温数据调整所述图像被输出的一色温。
所述校正色温数据的计算方式,如下式所示:
Figure PCTCN2019081324-appb-000004
其中,CT fnl为所述校正色温数据,CT 0为所述原始色温数据,α 0为一原始权重数据,CT i为所述对象色温数据,α i为所述权重数据,所述α 0与所有α i的总和为1。其中,所述α 0与所有α i可依事先采集的大量数据提取或由人为设定。
具体地,CT i可表示一图像中的不同记忆色对象的对象色温数据,例如:i=1相应于图像中的一肤色识别对象,i=2相应于图像中的一绿植识别对象,i=3相应于图像中的一太阳高度识别对象,但不以此为限。
具体地,所述图像色温校正装置还可被配置成为具有图像数据处理功能的装置的一部分,例如:笔记本电脑、平板电脑或智能手机中的一功能性模块,用以辅助处理有 关图像被输出前的色温校正功能。
此外,本发明的另一方面提供一种图像色温校正方法,所述方法可由一处理器执行,所述处理器可耦接一内存,所述内存可包括至少一程序指令,所述程序指令能由所述处理器执行,所述方法可包括步骤:由所述处理器计算一图像的一原始色温数据;由所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据;由所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据;及由所述处理器依据所述校正色温数据调整所述图像被输出的一色温;其中所述校正色温数据的计算方式,如下式所示:
Figure PCTCN2019081324-appb-000005
其中,CT fnl为所述校正色温数据,CT 0为所述原始色温数据,α 0为一原始权重数据,CT i为所述对象色温数据,α i为所述权重数据,所述α 0与所有α i的总和为1。例如:i=1相应于图像中的一肤色识别对象,i=2相应于图像中的一绿植识别对象,i=3相应于图像中的一太阳高度识别对象,但不以此为限。以下举例说明提取所述原始色温数据的实施方式,但不以此为限。
举例来说,所述原始色温数据可以依据一灰度世界算 法(Gray World Algorithm)从所述图像提取,所述灰度世界算法可以参照计算颜色恒常性算法的比较—第一部分:合成数据的方法和实验报告,如下式所示:
Figure PCTCN2019081324-appb-000006
其中,i为像素序号;ρ 1、ρ 2、ρ 3为红(r)、绿(g)、蓝(b)三通道数位值;n为像素总数。根据ρ,查询事先从测试实验获得的一ρ-CT表,可获得CT值作为所述原始色温数据。
替代地,所述原始色温数据还可利用下列方法步骤获得:计算图像像素的r、g、b值,将所述r、g、b值经伽马(gamma)变换转成R、G、B光学值(根据CIE转换公式),将R、G、B光学值经显示器的TM(transformation)矩阵转化成X、Y、Z三刺激值(根据CIE转换公式),将X、Y、Z三刺激值转化成CIE1931空间的x、y值(根据CIE转换公式),上述转换或变换的实现方式是本领域技术人员可以理解的;依据x、y值判断一色点所处色温分区(如图2所示,本领域技术人员可以理解的是,所述色温图也可呈现为彩色),依照不同色温分区进行计算,举例如下所示:
(一)若位于A区域,以色点P A为例,则:
基于一色温公式计算所述色点P A的色温(CT):
CT=-437*n^3+3601*n^2-6861*n+5514.31。
n=(x PA-0.332)/(y PA-0.1858),x PA、y PA为色点P A在图2中的x、y值。
累计至色温统计表(表1),以第i个色温CT(i)的统计值V CT为例,即V CT=V CT+1。
表1 色温统计表
色温 1000K 1500K …… CT(i) …… 15000K
统计值 V 1000 V 1500 …… V CT …… V 15000
(二)若位于B区域,以色点PB为例,则:
计算夹角α、β;
计算权重值γ=1-α/β;
累计权重值γ至色温统计表(表1)中15000K色温处,即V 15000=V 15000+γ。
(三)若位于C区域,则:
与B区域同理,累计权重值至色温统计表(表1)中1000K色温处,即V 1000=V 1000+γ。
结合权重计算图像色温,如下式所示:
Figure PCTCN2019081324-appb-000007
在一些实施例中,所述记忆色对象可选自由一肤色识别对象、一绿植识别对象或一太阳高度识别对象组成的一群组。
举例来说,当所述记忆色对象为所述肤色识别对象时,对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图(如以各种基于深度神经网络的自动调光的算法取得),提取在所述语义标签图中关于一 人像的掩膜,(如M person,如图3中的“人”的剖线区域),以所述掩膜对所述图像提取一人像分量(如P person=P*M person,其中P为初始图像),采用一常用肤色范围(如参照人脸检测中的肤色聚类报告)对所述人像分量(如P person)提取一肤色部分(如P skin),将所述肤色部分(如P skin)依据一色温统计结果产生一初始肤色色温(如CT skin),将所述初始肤色色温调整为一校正肤色色温(如CT corrskin)作为所述对象色温数据,所述校正肤色色温的计算方式如下式所示:
CT corrskin=CT m+CT skin-CT averskin
其中,CT corrskin为所述校正肤色色温,CT m为一中性色温常量(一般为4000K至5500K),CT skin为所述初始肤色色温,CT averskin为一肤色平均色温常量(可通过统计大量肤色图片计算所得)。
特别说明的是,采用肤色识别的原因是因为肤色属于记忆色,但是它本身处于暖色色调范围,为了避免肤色对统计色温的影响导致误判,所以将肤色单独提取,以便后期进行统计色温的校正。
另外,当所述记忆色对象为所述绿植识别对象时,对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于草及树的掩膜(如M green,如图3中的“草”与“树”的剖线区域),以所述掩膜对所述图像提取一绿植分量(如P green=P* M green,其中P为初始图像),将所述绿植分量(如P green)依据一色温统计结果产生一绿植色温(如CT green)作为所述对象色温数据。
特别说明的是,由于绿植的统计平均色温一般处于中性色温,所以不需要进行校正。加入绿植识别的原因是绿植相对于其余花卉等室外场景,属于记忆色,为了避免花卉等其余室外场景对色温的影响导致统计误判,所以将绿植部分单独提取,增加绿植影响权重,对统计色温进行校正。
另外,当所述记忆色对象为所述太阳高度识别对象时,以一太阳高度识别的神经网路算法(如各种从室外全景学习高动态范围的算法)对所述图像提取一太阳亮度(如I sun)及一太阳高度角(如y sun),所述太阳亮度的一取值为1或0,所述太阳高度角的一取值范围为0度至90度(°,degree;0°接近日出或日落,90°接近正午),将所述太阳亮度及所述太阳高度角与一太阳高度色温表(如下表2所示,是以赤道地区数据为例,其他地区数据需根据太阳高度角进行换算)比对产生一太阳高度色温(如CT sun)作为所述对象色温数据。
表2 太阳高度色温表
Figure PCTCN2019081324-appb-000008
Figure PCTCN2019081324-appb-000009
本发明的图像色温校正方法及装置实施例通过采用统计方法对图像色温进行初步估计,再采用目标识别法识别特殊物体(如不同记忆色对象)并计算色温,最后对估计色温进行校正,得到校正后的影像色温。
另一方面,如以现有的图像色温统计算法处理图像色温,在有大面积蓝色天空的夕阳图中,由于蓝色占图片的大部分,以统计结果为主则图像色温偏高(偏冷),但是在实际场景中,人们感受到的夕阳光源是偏低(偏暖);另,在冷光源下的人像图,由于肤色一般属于暖色,如果肤色占图像的大部分面积,则统计结果图像色温偏低(偏暖),但是在实际场景中,人们感受到的光源是偏高(偏冷);又,在有大面积单色红花的图像中,由于红色为暖色,统计结果图像色温偏低(偏暖),但是人们根据绿叶的记忆色可以判断场景光源为中性。
相较于单纯采用统计方法调整色温的现有技术,本发明的色温校正结果相应于图像所含的记忆色对象(如天空、 肤色、绿植等),可以获得提高影像色温估计的准确率,使色温校正结果更加接近人为观察结果等有益效果。
本发明已由上述相关实施例加以描述,然而上述实施例仅为实施本发明的范例。必需指出的是,已公开的实施例并未限制本发明的范围。相反地,包含于权利要求书的精神及范围的修改及均等设置均包括于本发明的范围内。

Claims (16)

  1. 一种图像色温校正方法,所述方法由一处理器执行,所述处理器耦接一内存,所述内存包括至少一程序指令,所述程序指令能由所述处理器执行,所述方法包括步骤:
    由所述处理器计算一图像的一原始色温数据;
    由所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据;
    由所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据;及
    由所述处理器依据所述校正色温数据调整所述图像被输出的一色温;
    其中所述校正色温数据的计算方式,如下式所示:
    Figure PCTCN2019081324-appb-100001
    其中,CT fnl为所述校正色温数据,CT 0为所述原始色温数据,α 0为一原始权重数据,CT i为所述对象色温数据,α i为所述权重数据,所述α 0与所有α i的总和为1。
  2. 如权利要求1所述的图像色温校正方法,其中,所述记忆色对象选自由一肤色识别对象、一绿植识别对象 或一太阳高度识别对象组成的一群组。
  3. 如权利要求2所述的图像色温校正方法,其中,当所述记忆色对象为所述肤色识别对象时,对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于一人像的掩膜,以所述掩膜对所述图像提取一人像分量,采用一常用肤色范围对所述人像分量提取一肤色部分,将所述肤色部分依据一色温统计结果产生一初始肤色色温,将所述初始肤色色温调整为一校正肤色色温作为所述对象色温数据,所述校正肤色色温的计算方式如下式所示:
    CT corrskin=CT m+CT skin-CT averskin
    其中,CT corrskin为所述校正肤色色温,CT m为一中性色温常量,CT skin为所述初始肤色色温,CT averskin为一肤色平均色温常量。
  4. 如权利要求2所述的图像色温校正方法,其中,当所述记忆色对象为所述绿植识别对象时,对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于草及树的掩膜,以所述掩膜对所述图像提取一绿植分量,将所述绿植分量依据一色温统计结果产生一绿植色温作为所述对象色温数据。
  5. 如权利要求2所述的图像色温校正方法,其中, 当所述记忆色对象为所述太阳高度识别对象时,以一太阳高度识别的神经网路算法对所述图像提取一太阳亮度及一太阳高度角,所述太阳亮度的一取值为1或0,所述太阳高度角的一取值范围为0度至90度,将所述太阳亮度及所述太阳高度角依据一太阳高度色温表查表产生一太阳高度色温作为所述对象色温数据。
  6. 一种图像色温校正装置,所述装置包括一处理器及一内存,所述内存耦接所述处理器且包括至少一程序指令,所述程序指令能由所述处理器执行,所述装置还包括:
    一统计色温模块,被配置成致使所述处理器计算一图像的一原始色温数据;
    一目标识别模块,被配置成致使所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据;及
    一色温校正模块,被配置成致使所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据,及依据所述校正色温数据调整所述图像被输出的一色温;
    其中所述校正色温数据的计算方式,如下式所示:
    Figure PCTCN2019081324-appb-100002
    其中,CT fnl为所述校正色温数据,CT 0为所述原始色温 数据,α 0为一原始权重数据,CT i为所述对象色温数据,α i为所述权重数据,所述α 0与所有α i的总和为1;
    所述目标识别模块包括一肤色识别单元,所述肤色识别单元被配置成致使所述处理器对于所述图像中的一肤色识别对象进行识别;
    所述目标识别模块包括一绿植识别单元,所述绿植识别单元被配置成致使所述处理器对于所述图像中的一绿植识别对象进行识别;及
    所述目标识别模块包括一太阳高度识别单元,所述太阳高度识别单元被配置成致使所述处理器对于所述图像中的一太阳高度识别对象进行识别。
  7. 如权利要求6所述的图像色温校正装置,其中,所述肤色识别单元被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于一人像的掩膜,以所述掩膜对所述图像提取一人像分量,采用一常用肤色范围对所述人像分量提取一肤色部分,将所述肤色部分依据一色温统计结果产生一初始肤色色温,将所述初始肤色色温调整为一校正肤色色温作为所述对象色温数据,所述校正肤色色温的计算方式如下式所示:
    CT corrskin=CT m+CT skin-CT averskin
    其中,CT corrskin为所述校正肤色色温,CT m为一中性色 温常量,CT skin为所述初始肤色色温,CT averskin为一肤色平均色温常量。
  8. 如权利要求6所述的图像色温校正装置,其中,所述绿植识别单元被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于草及树的掩膜,以所述掩膜对所述图像提取一绿植分量,将所述绿植分量依据一色温统计结果产生一绿植色温作为所述对象色温数据。
  9. 如权利要求6所述的图像色温校正装置,其中,所述太阳高度识别单元被配置成致使所述处理器以一太阳高度识别的神经网路算法对所述图像提取一太阳亮度及一太阳高度角,所述太阳亮度的一取值为1或0,所述太阳高度角的一取值范围为0度至90度,将所述太阳亮度及所述太阳高度角依据一太阳高度色温表查表产生一太阳高度色温作为所述对象色温数据。
  10. 一种图像色温校正装置,所述装置包括一处理器及一内存,所述内存耦接所述处理器且包括至少一程序指令,所述程序指令能由所述处理器执行,所述装置还包括:
    一统计色温模块,被配置成致使所述处理器计算一图像的一原始色温数据;
    一目标识别模块,被配置成致使所述处理器对于所述图像中的至少一种记忆色对象进行识别,并计算所述至少一种记忆色对象相应的至少一个对象色温数据;及
    一色温校正模块,被配置成致使所述处理器基于所述原始色温数据、所述至少一个对象色温数据及至少一个权重数据计算一校正色温数据,及依据所述校正色温数据调整所述图像被输出的一色温;
    其中所述校正色温数据的计算方式,如下式所示:
    Figure PCTCN2019081324-appb-100003
    其中,CT fnl为所述校正色温数据,CT 0为所述原始色温数据,α 0为一原始权重数据,CT i为所述对象色温数据,α i为所述权重数据,所述α 0与所有α i的总和为1。
  11. 如权利要求10所述的图像色温校正装置,其中,所述目标识别模块包括一肤色识别单元,所述肤色识别单元被配置成致使所述处理器对于所述图像中的一肤色识别对象进行识别。
  12. 如权利要求11所述的图像色温校正装置,其中,所述肤色识别单元被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于一人像的掩膜,以所述掩膜对所述图像提取一人像分量,采用一常用 肤色范围对所述人像分量提取一肤色部分,将所述肤色部分依据一色温统计结果产生一初始肤色色温,将所述初始肤色色温调整为一校正肤色色温作为所述对象色温数据,所述校正肤色色温的计算方式如下式所示:
    CT corrskin=CT m+CT skin-CT averskin
    其中,CT corrskin为所述校正肤色色温,CT m为一中性色温常量,CT skin为所述初始肤色色温,CT averskin为一肤色平均色温常量。
  13. 如权利要求10所述的图像色温校正装置,其中,所述目标识别模块包括一绿植识别单元,所述绿植识别单元被配置成致使所述处理器对于所述图像中的一绿植识别对象进行识别。
  14. 如权利要求13所述的图像色温校正装置,其中,所述绿植识别单元被配置成致使所述处理器对所述图像采用一场景区别及物体分割方法以产生含有多种掩膜的一语义标签图,提取在所述语义标签图中关于草及树的掩膜,以所述掩膜对所述图像提取一绿植分量,将所述绿植分量依据一色温统计结果产生一绿植色温作为所述对象色温数据。
  15. 如权利要求10所述的图像色温校正装置,其中,所述目标识别模块包括一太阳高度识别单元,所述太阳高度识别单元被配置成致使所述处理器对于所述图像中的 一太阳高度识别对象进行识别。
  16. 如权利要求15所述的图像色温校正装置,其中,所述太阳高度识别单元被配置成致使所述处理器以一太阳高度识别的神经网路算法对所述图像提取一太阳亮度及一太阳高度角,所述太阳亮度的一取值为1或0,所述太阳高度角的一取值范围为0度至90度,将所述太阳亮度及所述太阳高度角依据一太阳高度色温表查表产生一太阳高度色温作为所述对象色温数据。
PCT/CN2019/081324 2018-12-15 2019-04-03 图像色温校正方法及装置 WO2020118977A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811537822.8A CN109451292B (zh) 2018-12-15 2018-12-15 图像色温校正方法及装置
CN201811537822.8 2018-12-15

Publications (1)

Publication Number Publication Date
WO2020118977A1 true WO2020118977A1 (zh) 2020-06-18

Family

ID=65558832

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/081324 WO2020118977A1 (zh) 2018-12-15 2019-04-03 图像色温校正方法及装置

Country Status (2)

Country Link
CN (1) CN109451292B (zh)
WO (1) WO2020118977A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR3118517A1 (fr) * 2020-12-31 2022-07-01 L'oreal Procédé de détermination d’une couleur de la peau d’un visage et système correspondant
WO2022144233A1 (en) * 2020-12-31 2022-07-07 L'oreal Method for determining a skin colour of a face and corresponding system

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109451292B (zh) * 2018-12-15 2020-03-24 深圳市华星光电半导体显示技术有限公司 图像色温校正方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1443009A (zh) * 2002-02-20 2003-09-17 佳能株式会社 白平衡校正装置和白平衡校正方法
US20030222992A1 (en) * 2002-06-04 2003-12-04 Michael Kaplinsky Method for statistical analysis of images for automatic white balance of color channel gains for image sensors
CN103327345A (zh) * 2012-03-19 2013-09-25 华晶科技股份有限公司 白平衡处理方法及其处理装置
CN107959851A (zh) * 2017-12-25 2018-04-24 广东欧珀移动通信有限公司 色温检测方法及装置、计算机可读存储介质和计算机设备
CN109451292A (zh) * 2018-12-15 2019-03-08 深圳市华星光电半导体显示技术有限公司 图像色温校正方法及装置

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6788813B2 (en) * 2000-10-27 2004-09-07 Sony Corporation System and method for effectively performing a white balance operation
WO2006036027A1 (ja) * 2004-09-30 2006-04-06 Fujifilm Corporation 画像処理装置および方法、ならびに画像処理プログラム
US8717459B2 (en) * 2008-12-11 2014-05-06 Texas Instruments Incorporated Method and apparatus for improving automatic white balance with scene information
CN103051804B (zh) * 2012-12-28 2014-12-10 广东欧珀移动通信有限公司 一种移动终端智能拍照方法及系统
CN105100576B (zh) * 2014-04-21 2018-03-23 展讯通信(上海)有限公司 光源检测方法及装置
CN104320642B (zh) * 2014-10-11 2016-08-24 广东欧珀移动通信有限公司 一种图片处理的方法及装置
CN106973278B (zh) * 2014-11-11 2018-08-28 深圳瑞尔图像技术有限公司 一种参照人脸色彩特征的自动白平衡装置及方法
CN108229478B (zh) * 2017-06-30 2020-12-29 深圳市商汤科技有限公司 图像语义分割及训练方法和装置、电子设备、存储介质和程序
CN108282647B (zh) * 2018-01-31 2019-11-05 上海小蚁科技有限公司 色彩校正方法及装置、计算机可读存储介质、终端
CN108921040A (zh) * 2018-06-08 2018-11-30 Oppo广东移动通信有限公司 图像处理方法和装置、存储介质、电子设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1443009A (zh) * 2002-02-20 2003-09-17 佳能株式会社 白平衡校正装置和白平衡校正方法
US20030222992A1 (en) * 2002-06-04 2003-12-04 Michael Kaplinsky Method for statistical analysis of images for automatic white balance of color channel gains for image sensors
CN103327345A (zh) * 2012-03-19 2013-09-25 华晶科技股份有限公司 白平衡处理方法及其处理装置
CN107959851A (zh) * 2017-12-25 2018-04-24 广东欧珀移动通信有限公司 色温检测方法及装置、计算机可读存储介质和计算机设备
CN109451292A (zh) * 2018-12-15 2019-03-08 深圳市华星光电半导体显示技术有限公司 图像色温校正方法及装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR3118517A1 (fr) * 2020-12-31 2022-07-01 L'oreal Procédé de détermination d’une couleur de la peau d’un visage et système correspondant
WO2022144233A1 (en) * 2020-12-31 2022-07-07 L'oreal Method for determining a skin colour of a face and corresponding system

Also Published As

Publication number Publication date
CN109451292B (zh) 2020-03-24
CN109451292A (zh) 2019-03-08

Similar Documents

Publication Publication Date Title
US11323676B2 (en) Image white balance processing system and method
WO2020118977A1 (zh) 图像色温校正方法及装置
CN104182949B (zh) 基于直方图特征点配准的影像匀色与融合方法及系统
TWI532385B (zh) 白平衡處理方法及其處理裝置
CN106897981A (zh) 一种基于引导滤波的低照度图像增强方法
WO2020082593A1 (zh) 增强图像对比度的方法及其装置
CN105608677A (zh) 一种任意光线环境下的图像肤色美化方法及系统
CN106169182B (zh) 一种合成多幅不同曝光度图像的方法
WO2022116989A1 (zh) 图像处理方法、装置、设备和存储介质
WO2019001163A1 (zh) 一种白平衡调整方法、装置、相机及介质
CN108416754A (zh) 一种自动去除鬼影的多曝光图像融合方法
WO2019061766A1 (zh) 一种图像处理方法及装置
CN103854261B (zh) 色偏图像的校正方法
CN103714520A (zh) 基于fpga实现数字视频图像增强系统及其方法
WO2019134491A1 (en) Method and apparatus for processing image
WO2021218603A1 (zh) 图像处理方法及投影系统
CN105989583B (zh) 一种图像去雾方法
CN109712177A (zh) 图像处理方法、装置、电子设备和计算机可读存储介质
US20140334728A1 (en) Method and device of skin tone optimization in a color gamut mapping system
CN113206949A (zh) 基于熵加权图像梯度的半直接单目视觉slam方法
CN110852956A (zh) 一种高动态范围图像的增强方法
He et al. Single image dehazing with white balance correction and image decomposition
CN109658367A (zh) 基于彩色传递的图像融合方法
CN106686320B (zh) 一种基于数密度均衡的色调映射方法
CN107705263A (zh) 一种基于rgb‑ir传感器的自适应透雾方法和终端

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: 19897363

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: 19897363

Country of ref document: EP

Kind code of ref document: A1