Display device, control method, and program

Abstract

The appropriate luminous intensity of the backlight module is used for each type of content. [Solution] A display device comprising: a liquid crystal panel; an input means for inputting a first group of images including a first image; a backlight module capable of changing the luminescence brightness of a plurality of divided regions for irradiating the liquid crystal panel with light; a holding means for holding a plurality of trained models; and a light emission control means for controlling the luminescence brightness of the plurality of divided regions of the backlight module using one of the plurality of trained models based on the analysis of the first group of images.

Description

Technical Field

[0001] The present invention relates to a display device capable of changing the emission luminance of a backlight module.

Background Art

[0002] There is a demand for increasing the contrast of a display device that displays an image having a relatively wide dynamic range, such as an HDR (High Dynamic Range) image. Representative display devices include a liquid crystal display device (LCD (Liquid Crystal Display) device). In an LCD device, a liquid crystal panel adjusts the amount of transmitted light of light irradiated from a backlight module for each pixel. In an LCD device, since light irradiated from the backlight module cannot be completely blocked, black floating due to light leakage occurs. Therefore, in an LCD device, the contrast of the display tends to be lower than that of an OLED display device, which is a self-emitting display device.

[0003] When reducing black floating and improving the contrast in an LCD device, for example, a technique called local dimming is used. Local dimming is a technique for reducing black floating by controlling the emission luminance of the backlight module for each divided region. In conventional local dimming, for example, a specific rule is applied to the control value of the backlight module for each divided region so that luminance deficiency does not occur in a bright region of a small area. Patent Document 1 discloses a method of classifying bright regions detected from an image (for example, the sun, fire, etc.) and determining the luminance of the bright regions based on defined luminance values corresponding to the classification results.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, depending on the image, increasing the output of the backlight module to compensate for insufficient brightness can make black level distortion more noticeable, and there is sometimes a trade-off between insufficient brightness and suppressing black level distortion. The method disclosed in Patent Document 1 can be expected to increase brightness in bright areas, but there is a problem in that black level distortion becomes noticeable depending on the image.

[0006] Therefore, the present invention aims to use the appropriate luminous brightness of the backlight module for each type of content. [Means for solving the problem]

[0007] A display device comprising: a liquid crystal panel; input means for inputting a first group of images including a first image; a backlight module capable of changing the luminescence brightness of a plurality of divided regions for irradiating the liquid crystal panel with light; holding means for holding a plurality of trained models; and light emission control means for controlling the luminescence brightness of the plurality of divided regions of the backlight module using one of the plurality of trained models based on analysis of the first group of images. [Effects of the Invention]

[0008] According to the present invention, the luminous brightness of the backlight module can be set to be appropriate for each type of content. [Brief explanation of the drawing]

[0009] [Figure 1] This is a block diagram showing the functional blocks of a liquid crystal display device according to Example 1. [Figure 2] This figure shows an example of a divided region of the backlight according to Example 1. [Figure 3] This figure shows an example of a backlight brightness estimation point according to Example 1. [Figure 4] This figure shows an example of the analysis data related to Example 1. [Figure 5]This is a block diagram showing the functional blocks of the model learning unit according to Example 1. [Figure 6] This is a block diagram showing the functional block of the LD feature generation unit according to Example 1. [Figure 7] This figure shows an example of a notification message related to Example 1. [Figure 8] This is a block diagram showing the functional block of the image brightness adjustment unit according to Example 2. [Figure 9] This is a block diagram showing the functional blocks of the model learning unit according to Example 2. [Modes for carrying out the invention]

[0010] Embodiments of the present invention will be described below with reference to the drawings. The technical scope of the present invention is defined by the claims and is not limited by the embodiments illustrated below. Furthermore, not all combinations of features described in the embodiments are essential to the present invention. The contents described in this specification and drawings are illustrative and should not be considered limiting to the present invention. Various modifications are possible based on the spirit of the present invention, and these do not exclude them from the scope of the present invention. That is, all configurations combining each embodiment and its modified forms are included in the present invention.

[0011] [Example 1] The following describes Embodiment 1 of the present invention. Embodiment 1 shows a method for realizing the present invention in a liquid crystal display device 100 that achieves local dimming by backlight module control using a machine learning model.

[0012] Figure 1 is a block diagram showing the functional blocks of the liquid crystal display device 100 according to Embodiment 1. The liquid crystal display device 100 includes an image input / conversion unit 101, a local dimming control unit 102, a liquid crystal panel control unit 103, a liquid crystal panel 104, a backlight control unit 105, and a backlight module 106.

[0013] The image input / transformation unit 101 acquires image data (data of an image) from the outside. Specifically, the image input / transformation unit 101 has an input interface such as SDI (Serial Digital Interface), and inputs image data into the liquid crystal display device 100 from the outside via the input interface. Then, the image input / transformation unit 101 performs transformation processing such as gradation transformation and signal format transformation on the acquired (input) image data, and outputs the image data after the transformation processing as an input image.

[0014] The gradation transformation is, for example, gradation transformation using a one-dimensional look-up table (1D-LUT), and is gradation transformation according to the gamma value (panel gamma) of the liquid crystal panel 104. Here, consider the case where the gamma characteristic (the correspondence between the gradation value and the luminance; the gradation characteristic) of the image data acquired from the outside is a linear characteristic in which the luminance increases linearly with the increase of the gradation value, and the panel gamma is 2.0. In this case, gradation transformation using the inverse gamma of the panel gamma (that is, 1 / 2.0) is performed. Thereby, the acquired image data (image data having a linear characteristic) is converted into image data having a gamma characteristic in which the luminance is proportional to the 1 / 2.0 power of the gradation value. Note that the transformation processing in the image input / transformation unit 101 is not limited to gradation transformation using a 1D-LUT, and may include transformation processing using a three-dimensional look-up table (3D-LUT), gain adjustment, offset adjustment, matrix transformation, and the like.

[0015] The signal format transformation is, for example, processing for converting the signal format of the image data from YCbCr, XYZ, etc. to RGB. Note that the signal formats before and after the transformation are not limited to YCbCr, XYZ, and RGB.

[0016] The local dimming control unit 102 includes a preprocessing unit 10201, a learned model holding unit 10202, a backlight control value generation unit 10203, and a backlight luminance estimation unit 10204. Further, the local dimming control unit 102 is composed of a correction coefficient generation unit 10205, an image correction unit 10206, an analysis data generation unit 10207, an analysis data holding unit 10208, and a learned model selection unit 10209.

[0017] The preprocessing unit 10201 generates feature amounts (image feature amounts) based on the input image output from the image input / transformation unit 101. In the first embodiment, as shown in FIG. 2, the preprocessing unit 10201 generates image feature amounts by calculating the maximum gradation value and the average gradation value of the input image corresponding to a plurality of divided regions. That is, when the number of horizontal divisions of the backlight module 106 is n and the number of vertical divisions is m, the preprocessing unit 10201 generates two-dimensional data of n×m as the image feature amount. Then, the preprocessing unit 10201 outputs the generated image feature amount to the backlight control value generation unit 10203.

[0018] The learned model holding unit 10202 holds a model (learned model) learned to convert the image feature amount into a backlight control value. The backlight control value is a control value for controlling the backlight module 106. The learned model is a group of parameters applied to a neural network that converts the image feature amount into a backlight control value, and includes weights and biases between neurons constituting the neural network. The learned model held by the learned model holding unit 10202 is transmitted to the learned model selection unit 10209. A method for generating the learned model will be described later.

[0019] The backlight control value generation unit 10203 generates backlight control values ​​based on image features output from the preprocessing unit 10201. The backlight control values ​​are control values ​​for controlling the luminescence brightness of the backlight module 106 via the backlight control unit 105. As shown in Figure 2, the backlight module 106 has multiple divided regions that constitute the display surface, each of which has multiple light sources, and the luminescence brightness can be changed for each divided region. The light sources of the backlight module 106 are not particularly limited, but for example, they are LEDs (Light Emitting Diodes). The backlight control value generation unit 10203 generates backlight control values ​​for each divided region. The backlight control value generation unit 10203 then outputs the backlight control values ​​to the backlight brightness estimation unit 10204 and the backlight control unit 105.

[0020] In Example 1, the backlight control value generation unit 10203 is composed of at least a neural network that takes image features as input and outputs backlight control values. When the horizontal division of the backlight module 106 is n and the vertical division is m, the neural network of the backlight control value generation unit 10203 has a structure that inputs and outputs n × m two-dimensional data. Furthermore, the neural network of the backlight control value generation unit 10203 has a structure that can apply a trained model held by the trained model holding unit 10202. The trained model to be used can be switched by setting the trained model selected by the trained model selection unit 10209 (described later) to the neural network of the backlight control value generation unit 10203. Note that the network structure of the neural network of the backlight control value generation unit 10203 is not limited as long as it can generate backlight control values.

[0021] The backlight control value generation unit 10203 may include a configuration that generates backlight control values ​​based on specific rules in addition to a neural network. Furthermore, the backlight control value generation unit 10203 may include multiple neural networks or specific rules for generating backlight control values. In this case, the backlight control value generation unit 10203 can switch between a neural network and specific rules for generating backlight control values.

[0022] The backlight brightness estimation unit 10204 performs an estimation calculation of the brightness of the light irradiated from the backlight module 106 onto the liquid crystal panel 104 (backlight brightness) based on the backlight control value output from the backlight control value generation unit 10203. The backlight brightness estimation calculation estimates (calculates) the backlight brightness based on the backlight control value of each light source (each divided region) and the brightness distribution model of the light emitted from the light source (the part of the backlight module 106 corresponding to the divided region). Here, the backlight brightness estimation unit 10203 estimates (calculates) the backlight brightness for each brightness estimation point discretely arranged within the display surface as shown in Figure 3(A). Specifically, according to the following formula (1), the backlight control value B ij and weight W ij The backlight brightness L is calculated by performing a sum-of-products operation. In the following formula (1), n ​​represents the number of horizontal divisions of the backlight module, and m represents the number of vertical divisions of the backlight module. Also, B ij The backlight control values ​​are for vertical position i and horizontal position j, W ij This is the backlight control value B ij The weight applied, L, represents the backlight brightness at the brightness estimation point. The backlight lit in each divided region decreases in intensity as the distance increases. Therefore, the backlight control value B ij Weight W applied to it ij The closer the distance from the brightness estimation point, the greater the value.

[0023]

number

[0024] In the example shown in Figure 3(A), one luminance estimation point is placed for one divided region of the backlight module 106, but this is not limited to this. For example, they may be placed at the four corners of the divided region as shown in Figure 3(B), or one may be placed for each of the four divided regions as shown in Figure 3(C). The backlight luminance estimation unit 10204 interpolates the backlight luminance between the luminance estimation points and scales it to the resolution of the input image. Therefore, when the luminance estimation points are placed as shown in Figure 3(B), the density of luminance estimation points is higher than in Figure 3(A), suppressing interpolation errors, but increasing the computational load. On the other hand, when the luminance estimation points are placed as shown in Figure 3(C), the density of luminance estimation points is lower than in Figure 3(A), suppressing computational load, but increasing interpolation errors. The method for scaling the backlight luminance can be, for example, bicubic interpolation or bilinear interpolation, but it is not limited to these methods as long as it is possible to scale it to the resolution of the input image. The backlight luminance estimation unit 10204 then outputs the estimated backlight luminance to the correction coefficient generation unit 10205. Alternatively, the backlight brightness estimation unit 10204 may output discretely arranged backlight brightness values ​​to the correction coefficient generation unit 10205 without scaling the calculated backlight brightness, as shown in Figure 3(A). In this case, the correction coefficient generation unit 10205 scales the calculated correction coefficient to the resolution of the input image.

[0025] The correction coefficient generation unit 10205 generates a correction coefficient to be applied to the input image output from the image input / conversion unit 101, based on the backlight brightness output from the backlight brightness estimation unit 10204. In Example 1, the correction coefficient Gt is calculated using the reciprocal of the backlight brightness L (a value normalized to 0.0 to 1.0) according to the following formula (2). For example, if the backlight brightness is reduced to 1 / 3, the reduction in backlight brightness can be compensated for by the image by increasing the input image by its reciprocal, i.e., by 3 times. Note that the method for calculating the correction coefficient is not limited to the reciprocal of the backlight brightness.

[0026]

number

[0027] The image correction unit 10206 generates (calculates) the pixel values ​​of the corrected image by multiplying the correction coefficient output from the correction coefficient generation unit 10205 by the pixel values ​​of the input image output from the image input / conversion unit 101. In Example 1, the RGB values ​​(Vrc, Vgc, Vbc), which are the pixel values ​​of the corrected image, are generated by multiplying the RGB values ​​(R, G, B) = (Vr, Vg, Vb), which are the pixel values ​​of the input image, by the correction coefficient Gt corresponding to that pixel, according to the following formula (3). The image correction unit 10206 then outputs the generated corrected image to the liquid crystal panel control unit 103. Formula (3) Vrc = Vr × Gt Vgc = Vg × Gt Vbc = Vb × Gt The analysis data generation unit 10207 determines whether the input image output from the image input / conversion unit 101 satisfies specific conditions, and outputs the determination result as input image analysis data to the analysis data holding unit 10208, which will be described later.

[0028] Specific conditions are those used to predict whether local dimming is likely to cause saturation, black level issues, or halos. A condition for predicting saturation is whether there are many pixels with high tonal values ​​in the image. On the other hand, a condition for predicting black level issues is whether there are many pixels with low tonal values ​​in the image.

[0029] Below, we will explain an example of a specific discrimination method, assuming that Condition 1 is "whether there are many pixels with high tonal values ​​in the image," Condition 2 is "whether there are many pixels with low tonal values ​​in the image," and Condition 3 is "whether halos are likely to occur."

[0030] In condition 1, if the number of pixels in the image with a grayscale value greater than S1 is greater than the threshold N1, the analysis data generation unit 10207 determines that the image satisfies condition 1.

[0031] In condition 2, if the number of pixels in the image with a grayscale value smaller than S2 is greater than the threshold N2, the analysis data generation unit 10207 determines that the image satisfies condition 2.

[0032] In condition 3, if the number of pixels in the image with a grayscale value greater than S3 is greater than the threshold N3, and the number of pixels with a grayscale value less than S4 is greater than the threshold N4, the analysis data generation unit 10207 determines that the image satisfies condition 3.

[0033] The analysis data generation unit 10207 transmits the determination results of conditions 1 to 3, determined by the method described above, to the analysis data holding unit 10209 as input image analysis data. The specific conditions described above are just examples, and other conditions may be applied.

[0034] Once the analysis data generation unit 10207 has finished sending the input image analysis data to the analysis data holding unit 10208, it sends a notification to the trained model selection unit 10209 indicating that processing is complete.

[0035] The analysis data holding unit 10208 receives input image analysis data from the analysis data generation unit 10207 and updates the input image group analysis data, which is generated using frames prior to the current frame and is held by the analysis data holding unit 10208, by adding the input image analysis data.

[0036] Here, the input image set analysis data is a set of values ​​that counts how many images satisfy conditions 1, 2, and 3 across several frames of the input image. If NUM1 is the number of images that satisfy condition 1 before the update, NUM2 is the number of images that satisfy condition 2, and NUM3 is the number of images that satisfy condition 3, then the input image set analysis data before the update will be [NUM1, NUM2, NUM3]. In contrast, if the input image analysis data input to the analysis data storage unit 10208 satisfies conditions 1 and 2 but does not satisfy condition 3, then the input image set analysis data after the update will be [NUM1+1, NUM2+1, NUM3]. The analysis data storage unit 10208 transmits the updated input image set analysis data to the trained model selection unit 10209.

[0037] Furthermore, an upper limit may be set on the number of images to be used as input image group analysis data. In that case, it is necessary to separately maintain input image analysis data for each frame used to calculate the input image group analysis data up to the upper limit of the number of images, and the oldest input image analysis data is deleted and the latest input image analysis data is added to the set of retained data. The input image group analysis data is calculated each time by adding up the input image analysis data for each frame up to the upper limit of the number of images, according to the conditions.

[0038] The trained model selection unit 10209 selects a trained model to transmit to the backlight control value generation unit 10203 based on the input image group analysis data received from the analysis data holding unit 10208 and the trained models held in the trained model holding unit 10202.

[0039] As a prerequisite, the trained model holder 10202 holds multiple trained models, and the training conditions, such as the images (training images) input to the model training unit 200 in the trained model generation process described later, differ between each trained model. Furthermore, each trained model is associated with training image group analysis data for the images used during training, and the method for generating the training image group analysis data is the same as the method for generating the input image group analysis data described above. Details regarding the training method will be described later.

[0040] The trained model selection unit 10209 searches for and selects the training image analysis data that is closest to the input image analysis data from each training image analysis data set, and transmits the trained model associated with the selected training image analysis data to the backlight control value generation unit 10203.

[0041] An example of a specific selection method is described below. Figure 4(A) is an example of input image set analysis data. In Figure 4(A), there are 80 images that satisfy condition 1, 20 images that satisfy condition 2, and 15 images that satisfy condition 3.

[0042] Furthermore, Figure 4(B) shows an example of each trained model held by the trained model holder 10202 and the associated trained image set analysis data. In Figure 4(B), for example, trained model 2 has 50 images that satisfy condition 1, 50 images that satisfy condition 2, and 25 images that satisfy condition 3. When each analysis data is in the state shown in the tables in Figure 4, the similarity D of each trained image set analysis data to the input image set analysis data can be calculated using formula (4).

[0043]

number

[0044] The liquid crystal panel control unit 103 controls the transmittance of the liquid crystal panel 104 (transmittance distribution within the display surface) based on the corrected image output from the image correction unit 10206, so that the image based on the corrected image is displayed on the liquid crystal panel 104. The liquid crystal panel control unit 103 is an example of a display control means.

[0045] The liquid crystal panel 104 is controlled by the liquid crystal panel control unit 103 and displays an image on its display surface.

[0046] The backlight control unit 105 controls the luminescence brightness of the backlight module 106 (the light source of the backlight module 106) according to the backlight control value output from the backlight control value generation unit 10203. The backlight control unit 105 is an example of a light emission control means. For example, the backlight control unit 105 determines the duty cycle of PWM (Pulse Width Modulation) control according to the backlight control value, and controls the luminescence brightness of the backlight module 106 by PWM control at the determined duty cycle.

[0047] The backlight module 106 illuminates the back of the liquid crystal panel 104 with light. As described above, the luminescence brightness of the backlight module 106 is adjustable. In Embodiment 1, multiple divided regions constituting the display surface are pre-set, and the backlight module 106 has multiple light sources corresponding to each of the multiple divided regions, and the luminescence brightness can be changed for each divided region.

[0048] As explained with reference to Figure 1, the liquid crystal display device 100 can control the backlight module 106 by applying a trained model selected based on input image group analysis data to the neural network of the backlight control value generation unit 10203.

[0049] Figure 5 is a block diagram showing the functional blocks of the model learning unit 200 according to Embodiment 1. The model learning unit 200 includes an image input / conversion unit 201, a preprocessing unit 202, and a backlight control value generation unit 203. The model learning unit 200 also includes a first setting unit 204, an LD feature generation unit 205, a second setting unit 206, a target feature generation unit 207, an error calculation unit 208, a trained model output unit 209, an analysis data generation unit 210, and an analysis data holding unit 211. The model learning unit 200 is, for example, a program that runs on a computer, but is not limited to that, and may also operate as a functional block constituting the liquid crystal display device 100.

[0050] The image input / conversion unit 201 acquires image data (image data) from an external source. Specifically, the image input / conversion unit 201 acquires image data from a group of files stored in the computer's memory, but is not limited to this; image data may also be input via an input interface such as SDI. The image input / conversion unit 201 can then perform the same conversion processing on the acquired (input) image data as the image input / conversion unit 101 of the liquid crystal display device 100, and outputs the converted image data as the input image. However, the image input / conversion unit 201 may also acquire (input) pre-converted image data. In this case, the image input / conversion unit 201 does not need to perform image data conversion processing.

[0051] The preprocessor 202 generates feature quantities (image feature quantities) based on the input image output from the image input / conversion unit 201. Specifically, the preprocessor 202 generates image feature quantities in the same way as the preprocessor 10201 of the liquid crystal display device 100. That is, if the number of horizontal divisions of the backlight module 106 of the liquid crystal display device 100 is n and the number of vertical divisions is m, the preprocessor 202 generates n × m two-dimensional data as image feature quantities. The preprocessor 202 then outputs the generated image feature quantities to the backlight control value generation unit 203, the LD feature quantity generation unit 205, and the target feature quantity generation unit 207, respectively.

[0052] The backlight control value generation unit 203 is composed of a neural network that outputs backlight control values ​​based on image features output from the preprocessing unit 202. The neural network of the backlight control value generation unit 203 has the same structure as the neural network of the backlight control value generation unit 10203 of the liquid crystal display device 100. If the horizontal division number of the backlight module 106 of the liquid crystal display device 100 is n and the vertical division number is m, the neural network of the backlight control value generation unit 203 has a structure that inputs and outputs n × m two-dimensional data. The backlight control value generation unit 203 receives the parameters generated by the error calculation unit 208 and reflects them in the neural network. The backlight control value generation unit 203 then outputs the backlight control values ​​generated by the neural network to the LD feature generation unit 205. In addition, the backlight control value generation unit 203 outputs the parameters of the neural network generated by learning as a learned model to the learned model output unit 209.

[0053] The first setting unit 204 outputs parameters to the LD feature generation unit 205. Specifically, the first setting unit 204 outputs the panel contrast and the luminance distribution model to the LD feature generation unit 205. The panel contrast is the contrast value of the liquid crystal panel 104 of the liquid crystal display device 100. The luminance distribution model is a model of the luminance distribution of light emitted from the light source of the liquid crystal display device 100 (the part of the backlight module 106 corresponding to the divided area). The parameters set by the first setting unit 204 are obtained from the computer's memory, but are not limited to this, and may also be input from an external source via an OSD (On Screen Display) menu or the like. Note that the parameters set by the first setting unit 204 to the LD feature generation unit 205 are not limited to the panel contrast and the luminance distribution model, but may also include other parameters necessary for the LD feature generation unit 205.

[0054] The LD feature generation unit 205 generates LD (Local Dimming) features based on the image features output from the preprocessing unit 202 and the backlight control values ​​output from the backlight control value generation unit 203. In Example 1, the LD feature generation unit 205 generates LD features using parameters output from the first setting unit 204. The parameters output from the first setting unit 204 to the LD feature generation unit 205 are panel contrast and luminance distribution models, but are not limited to these; other parameters may be used, and external parameters may not be used at all. The LD features correspond to a simulated image of local dimming in the liquid crystal display device 100. The LD feature generation unit 205 then outputs the generated LD features to the error calculation unit 208. Details of the processing of the LD feature generation unit 205 will be described later.

[0055] The second setting unit 206 outputs parameters to the target feature generation unit 207. Specifically, the second setting unit 206 outputs the target contrast to the target feature generation unit 207. The target contrast is a contrast value used to convert the image features output from the preprocessing unit 202 to the target contrast. The parameters set by the second setting unit 206 are obtained from the computer's memory, but are not limited to this; they may also be input externally via an OSD (On Screen Display) menu or the like. Note that the parameters set by the second setting unit 206 to the target feature generation unit 207 are not limited to the target contrast and may include other parameters necessary for the target feature generation unit 207.

[0056] The target feature generation unit 207 generates target features based on the image features output from the preprocessing unit 202. The target features correspond to the target display image when local dimming control is performed on the liquid crystal display device 100. In Embodiment 1, the target feature generation unit 207 generates target features by converting the image features into a target contrast output from the second setting unit 206. Specifically, the data value Vt of the target feature data is calculated by performing a gain / offset calculation on the image feature data value V based on the target contrast Ct, according to the following formula (5). The target contrast is obtained from the second setting unit 206. If the contrast ratio of the target contrast is 1,000,000:1, then 1,000,000 is entered for the target contrast Ct.

[0057]

number

[0058] In Example 1, the target feature generation unit 207 generates target features by converting image features to a target contrast, but is not limited to this; any target feature corresponding to the target display image when local dimming control is performed on the liquid crystal display device 100 is acceptable. For example, the target feature generation unit 207 may acquire pre-prepared target features from an external source. Alternatively, the target feature generation unit 207 may output the image features output from the preprocessing unit 202 as the target features. The target feature generation unit 207 then outputs the generated target features to the error calculation unit 208.

[0059] The error calculation unit 208 generates the parameters for the neural network of the backlight control value generation unit 203 based on the difference between the LD features output from the LD feature generation unit 205 and the target features output from the target feature generation unit 207. The difference between the LD features and the target features is, for example, L1 (the sum of the absolute values ​​of the differences between each pixel). In this case, the error calculation unit 208 calculates the parameters for the neural network of the backlight control value generation unit 203 so as to minimize the difference between the LD features and the target features. The backlight control value generation unit 203 then outputs the generated parameters to itself.

[0060] The method for generating parameters by the error calculation unit 208 is not limited to one based on the difference between the LD feature and the target feature, but may be based on other methods. For example, the error calculation unit 208 may generate the parameters of the neural network of the backlight control value generation unit 203 based on the truth value determination of the LD feature and the target feature by the classifier. The classifier is composed of a neural network. In this case, the error calculation unit 208 is trained so that when a combination of image feature and target feature is input, the output value approaches 1 (true), and when a combination of image feature and LD feature is input, the output value approaches 0 (fake). Furthermore, the error calculation unit 208 calculates the parameters of the neural network of the backlight control value generation unit 203 so that the output value of the classifier when a combination of image feature and LD feature is input approaches 1 (true).

[0061] Furthermore, the parameter generation method by the error calculation unit 208 may be a combination of multiple methods. For example, a method based on the difference between the LD feature and the target feature may be combined with a method based on the truth value determination of the LD feature and the target feature by the classifier.

[0062] The trained model output unit 209 associates the trained model, which is the neural network parameter of the backlight control value generation unit 203, with the training image group analysis data held by the analysis data holding unit 221 (described later), and then outputs it externally. For example, the trained model output unit 209 outputs the trained model to the trained model holding unit 10202 of the liquid crystal display device 100 via external memory or the like.

[0063] The analysis data generation unit 210 determines whether the input image (training image) output from the image input / conversion unit 201 satisfies specific conditions, and outputs the determination result as training image analysis data to the analysis data holding unit 211, which will be described later.

[0064] The method for generating the training image analysis data is the same as that of the analysis data generation unit 10207 in Figure 1 above. Therefore, a detailed explanation of the analysis data generation method will be omitted.

[0065] The analysis data holding unit 211 receives the training image analysis data from the analysis data generation unit 210 and updates the training image group analysis data, which is held in the analysis data holding unit 211 and generated from at least frames prior to the current frame, by adding the training image analysis data.

[0066] The specific method for updating the training image set analysis data is the same as that of the analysis data storage unit 10208 in Figure 1, so the explanation will be omitted.

[0067] Furthermore, an upper limit may be set on the number of images to be retained as training image set analysis data. In this case, the method for updating the training image set analysis data is the same as that of the analysis data retention unit 10208 described above, so the explanation will be omitted.

[0068] As explained with reference to Figure 5, the model learning unit 200 of Example 1 can be trained to output a backlight control value so that the displayed image due to local dimming approaches the target value.

[0069] Figure 6 is a block diagram showing the functional blocks of the LD feature generation unit 205 according to Example 1. The LD feature generation unit 205 includes a backlight brightness estimation unit 20501, a correction coefficient generation unit 20502, a correction unit 20503, a contrast conversion unit 20504, and a synthesis unit 20505.

[0070] The backlight brightness estimation unit 20501 estimates the brightness of the light illuminating the liquid crystal panel 104 when the backlight module 106 of the liquid crystal display device 100 is illuminated using the backlight control value output from the backlight control value generation unit 203 (backlight brightness). Similar to the backlight brightness estimation unit 10204, the estimation calculation of the backlight brightness estimation unit 20501 estimates (calculates) the backlight brightness for each brightness estimation point discretely arranged within the display surface, as shown in Figure 3(A), based on the backlight control value and the brightness distribution model. Here, the backlight brightness estimation unit 20501 estimates (calculates) one backlight brightness for each divided region of one backlight module. The backlight brightness estimation unit 20501 obtains the brightness distribution model used for the estimation calculation from the first setting unit 204. The backlight brightness estimation unit 20501 then outputs the estimated backlight brightness to the correction coefficient generation unit 20502 and the synthesis unit 20505.

[0071] The correction coefficient generation unit 20502 generates a correction coefficient to be applied to the image features based on the backlight brightness output from the backlight brightness estimation unit 20501. The method for generating the correction coefficient by the correction coefficient generation unit 20502 is the same as that of the correction coefficient generation unit 10205 of the liquid crystal display device 100, for example, by calculating it using the reciprocal of the backlight brightness. The correction coefficient generation unit 20502 then outputs the generated correction coefficient to the correction unit 20503.

[0072] The correction unit 20503 generates (calculates) corrected features by multiplying the image features output from the preprocessing unit 202 by the correction coefficient output from the correction coefficient generation unit 20502. Specifically, the data value Vc of the corrected features is calculated by multiplying the data value V of the image features by the correction coefficient Gt according to the following formula (6). Here, the data value V of the features is a normalized value between 0.0 and 1.0. Formula (6) Vc = V × Gt The correction unit 20503 then outputs the generated correction feature quantity to the contrast conversion unit 20504.

[0073] The contrast conversion unit 20504 generates panel features by converting the correction features output from the correction unit 20503 into panel contrast. The panel features correspond to a simulated image when the backlight module 106 of the liquid crystal display device 100 is made to emit light uniformly and the correction features output from the correction unit 20503 are displayed on the liquid crystal panel 104. Specifically, the data value Vp of the panel features is calculated by performing a gain / offset calculation on the data value Vc of the correction features based on the panel contrast Cp, according to the following formula (7). The panel contrast is the contrast value of the liquid crystal panel 104 of the liquid crystal display device 100 and is obtained from the first setting unit 204. If the contrast of the liquid crystal panel 104 of the liquid crystal display device 100 is 1000:1, then 1000 is entered for the panel contrast Cp.

[0074]

number

[0075] The contrast conversion unit 20504 then outputs the corrected feature quantities (panel feature quantities) converted to panel contrast to the synthesis unit 20505.

[0076] The synthesis unit 20505 generates LD (Local Dimming) features by multiplying the panel features output from the contrast conversion unit 20504 by the backlight brightness output from the backlight brightness estimation unit 20501. Specifically, the data value Vl of the LD features is calculated by multiplying the data value Vp of the panel features by the backlight brightness L according to the following formula (8). Formula (8) Vl = Vp × L As explained with reference to Figure 6, the LD feature generation unit 205 of Example 1 can generate data equivalent to a simulation image of local dimming for calculating the error with the target feature.

[0077] By generating a trained model using the method described in this Example 1 and selecting a trained model from multiple candidates based on the input image analysis data, it is possible to achieve local dimming image quality that is more suitable for the content. Specifically, in content containing multiple images with many dark areas, a trained model that suppresses black level lifting in dark areas will be applied, and in content containing many images with many bright areas, a trained model that prioritizes brightness enhancement in bright areas will be applied. Furthermore, in content containing many images prone to halos, a trained model that suppresses halos will be applied.

[0078] In this embodiment, when the input content of the model learning unit 200 contains many dark images, the dark areas are more numerous than the bright areas, making it easier to reflect errors in the dark areas and generating a trained model with a high black level suppression effect. On the other hand, when the input content of the model learning unit 200 contains many bright images, the bright areas are more numerous than the dark areas, making it easier to reflect errors in the bright areas and generating a trained model that prioritizes high brightness. Furthermore, when the input content of the model learning unit 200 contains many images that tend to produce halos when local dimming is performed, the errors in the halo regions are more likely to be reflected, generating a trained model with a high halo suppression effect.

[0079] As mentioned earlier, there is a trade-off between suppressing black levels (or halos) and prioritizing high brightness. Therefore, applying a high-brightness-prioritizing trained model to content that contains many dark images, for example, can easily result in black levels being suppressed.

[0080] As described in this embodiment, the trained model selection unit 10209 of the liquid crystal display device 100 selects a trained model associated with the trained image group analysis data that is closest to the input image group analysis data representing the characteristics of the input content. Therefore, a trained model with a high black level suppression effect can be applied to content that contains many dark images. Similarly, if content containing many bright images is input, a high-brightness priority trained model is applied, and if content containing many images prone to halos is input, a trained model with a high halo suppression effect is applied. As a result, it becomes possible to achieve light source control that suppresses image quality degradation due to local dimming for each input content.

[0081] Furthermore, while each preprocessor unit 10201 and 202 generated image features based on the number of divisions of the backlight module 106, this is not limited to this. Image features with the same resolution as the input image may also be generated. In that case, each subsequent backlight control value generation unit 10203 and 203 uses a neural network that takes image features with the same resolution as the input image as input and generates a backlight control value composed of two-dimensional data with the number of divisions (n ​​× m) of the backlight module. In this way, each preprocessor unit 10201 and 202 preserves as much of the pixel data of the input image as possible, making it possible to generate backlight control values ​​that are more adapted to the characteristics of each image.

[0082] Furthermore, image features with the same resolution as the input image may be used to generate LD features in the LD feature generation unit 205 and to generate target features in the target feature generation unit 207. Doing so improves the accuracy of the errors calculated by each subsequent error calculation unit 208.

[0083] Furthermore, although it has been stated that the trained model selection unit 10209 calculates the similarity of multiple trained models stored in the trained model holding unit 10202 and transmits the trained model with the highest similarity to the backlight control value generation unit 10203, this method is not limited to that described above. If the similarity of each trained model falls below a pre-set similarity threshold, the trained model selection unit 10209 may prompt the user of the liquid crystal display device 100 to retrain the model using the current input image.

[0084] For example, in the aforementioned example, the similarity scores for each trained model were 2.9 for trained model 1, 1.6 for trained model 2, 0.5 for trained model 3, and 0.7 for trained model 4. If the similarity threshold is 3.0, the trained model selection unit 10209 checks whether the similarity score for all trained models is above 3.0. In this example, since the similarity score for each trained model is below 3.0, the trained model selection unit 10209 uses a notification unit (not shown) to notify the user that the model will be retrained using the current input image. This notification unit (not shown) may or may not be a mechanism that instructs the liquid crystal panel control unit 103 to display a notification message on the liquid crystal panel 104 of the liquid crystal display device 100. An example of a notification message shown to the user is shown in Figure 7. As shown in Figure 7, this message may also be superimposed as a pop-up message on a part of the content being viewed.

[0085] This retraining notification function enables backlight control that minimizes black level issues and brightness deficiencies for a wider variety of input images.

[0086] Furthermore, when generating analysis data in each analysis data generation unit 10207 and 210, metadata of the input image may be used. In this embodiment, an example was described in which the analysis data generation units 10207 and 210 determine that an image satisfies condition 1 if the number of pixels in the image with a grayscale value greater than S1 is greater than the threshold N1. For example, if the input image of the liquid crystal display device 100 is HDR10+ or ​​Dolby Vision, dynamic metadata is attached to the content, and brightness information for each frame is added to the content. Each analysis data generation unit 10207 and 210 may determine whether each image satisfies condition 1 based on whether this brightness information for each frame is greater than a pre-set threshold.

[0087] Furthermore, if the trained model used by the backlight control value generation unit 10203 switches while the user is viewing content, the shock caused by the trained model switch (such as image distortion) may be visible to the user. To solve this problem, a mechanism to control the timing of the trained model switch may be added.

[0088] For example, an input switching detection unit (not shown) may be added before the analysis data generation unit 10207 of the liquid crystal display device 100. This input switching detection unit detects when the input content of the liquid crystal display device 100 is switched, and when a switch is detected, a reset unit (not shown) deletes the input image group analysis data held by the analysis data holding unit 10208. Furthermore, it instructs the analysis data generation unit 10207 to read the input content before the user plays the content and generate input image group analysis data for this input content. Furthermore, the trained model selection unit 10209 selects a trained model from the trained model holding unit 10202 based on the input image group analysis data and transmits the selected trained model to the backlight control value generation unit 10203.

[0089] In this way, users can view content with backlight control using a trained model appropriate for the input content, without experiencing any visual shock from switching between trained models.

[0090] The reset unit described above operates when it detects a change in the input content of the liquid crystal display device 100, but this is not limited to that. The reset unit may also operate when the liquid crystal display device 100 receives a request from the user to reset the input image analysis data via a remote control (not shown) or the like.

[0091] Furthermore, as a method for switching between learned models, the learned model selection unit 10209 may not transmit the change in learned model to the backlight control value generation unit 10203, but instead instruct the liquid crystal panel control unit 103 to notify the user of the recommendation to change as a pop-up message. A specific example of a pop-up message has been described above, so a detailed explanation will be omitted. When the liquid crystal display device 100 receives a control signal from the user allowing a change in the learned model via a receiving unit (not shown), the receiving unit instructs the learned model selection unit 10209 to output a model change notification to the next stage.

[0092] Furthermore, while we have explained how to display pop-up notifications to users regarding the retraining of trained models and changes to the model, these are not the only notification methods available.

[0093] A communication unit (not shown) within the liquid crystal display device 100 may notify the user's communication device in accordance with the user's communication device information that has been previously registered in the liquid crystal display device 100.

[0094] [Example 2] Example 1 describes a method for achieving local dimming tailored to image characteristics by selecting a trained model that generates backlight control values ​​based on input image group analysis data of the liquid crystal display device 100.

[0095] Example 2 focuses on image correction to bring the brightness of image data closer to the brightness of a target image, and describes a method for generating these image correction values ​​from a trained model. For example, we propose a mechanism that allows changing the image correction value based on the characteristics of the input image by creating target images that increase contrast in brighter scenes and improve gradation in darker scenes, and then creating trained models for each.

[0096] Figure 8 is a block diagram showing the functional blocks of the image brightness adjustment unit 302 according to Example 2.

[0097] The image brightness adjustment unit 302 comprises a preprocessing unit 30201, a trained model holding unit 30202, an image correction value generation unit 30203, an image correction unit 30204, an analysis data generation unit 30205, an analysis data holding unit 30206, and a trained model selection unit 30207. In Embodiment 2, the image brightness adjustment unit 302 is described as a program that runs on a computer, but it is not limited to this and may operate as a functional block constituting a liquid crystal display device (not shown).

[0098] The preprocessing unit 30201 generates feature quantities (image feature quantities) based on the input image of the image brightness adjustment unit 302. In Embodiment 2, the preprocessing unit 30201 converts the input image into image feature quantities that indicate brightness. For example, the RGB values ​​(R value, G value, B value) = (Vr, Vg, Vb), which are the pixel values ​​of the input image, may be converted to Y using formula (9). Formula (9) Y = 0.299Vr + 0.587Vg + 0.114Vb The trained model storage unit 30202 stores a model (trained model) that has been trained to convert image features into image correction values. The image correction values ​​are correction values ​​used by the image correction unit 30204, described later, to correct the brightness of the input image. For example, the image correction values ​​are two-dimensional data to which different gain values ​​can be applied to each pixel of the input image. The trained model is a set of parameters applied to the neural network that converts image features into image correction values, and includes weights and biases between neurons that make up the neural network. The trained model stored in the trained model storage unit 30202 is transmitted to the trained model selection unit 30207. The method for generating the trained model will be described later.

[0099] The image correction value generation unit 30203 generates image correction values ​​based on the image features output from the preprocessing unit 30201.

[0100] In Example 2, the image correction value generation unit 30203 is comprised of at least a neural network that takes image features as input and outputs image correction values. When the number of horizontal pixels in the input image is n and the number of vertical pixels is m, the neural network of the image correction value generation unit 30203 has a structure that inputs and outputs n × m two-dimensional data. Furthermore, the neural network of the image correction value generation unit 30203 has a structure that allows the application of a trained model held by the trained model holding unit 30202. The trained model used can be switched by setting the trained model selected by the trained model selection unit 30207 (described later) to the neural network of the image correction value generation unit 30203. Note that the network structure of the neural network of the image correction value generation unit 30203 is not limited as long as it can generate image correction values.

[0101] The image correction value generation unit 30203 may include a configuration that generates image correction values ​​based on specific rules in addition to a neural network. Furthermore, the image correction value generation unit 30203 may include multiple neural networks or specific rules for generating image correction values. In this case, the image correction value generation unit 30203 can switch between a neural network and specific rules for generating image correction values.

[0102] The image correction unit 30204 generates (calculates) the pixel values ​​of the corrected image by multiplying the image correction values ​​output from the image correction value generation unit 30203 by the pixel values ​​of the input image to the image brightness adjustment unit 302. In Example 2, according to the following formula (10), the RGB values ​​(R value, G value, B value) = (Vr, Vg, Vb), which are the pixel values ​​of the input image, are multiplied by the image correction value Gt corresponding to that pixel to generate the RGB values ​​(Vrc, Vgc, Vbc), which are the pixel values ​​of the corrected image. The image correction unit 30204 then inputs the generated corrected image to a subsequent liquid crystal display device (not shown). Formula (10) Vrc = Vr × Gt Vgc = Vg × Gt Vbc = Vb × Gt The analysis data generation unit 30205 determines whether the input image of the image brightness adjustment unit 302 satisfies specific conditions, and outputs the determination result as input image analysis data to the analysis data holding unit 30206, which will be described later.

[0103] Specific conditions include, for example, whether the input content consists mostly of dark scenes or mostly of bright scenes.

[0104] Below, we will explain an example of a specific method for determining whether condition 1 is "whether there are many pixels with high grayscale values ​​in the image" or condition 2 is "whether there are many pixels with low grayscale values ​​in the image".

[0105] Under condition 1, if the number of pixels in the image with a grayscale value greater than S1 is greater than the threshold N1, the analysis data generation unit 30205 determines that the image satisfies condition 1.

[0106] In condition 2, if the number of pixels in the image with a grayscale value smaller than S2 is greater than the threshold N2, the analysis data generation unit 30205 determines that the image satisfies condition 2.

[0107] The analysis data generation unit 30205 transmits the determination results of conditions 1 and 2, determined by the method described above, as input image analysis data to the analysis data holding unit 30206. The specific conditions described above are just examples, and other conditions may be applied.

[0108] Once the analysis data generation unit 30205 has finished transmitting the input image analysis data to the analysis data holding unit 30206, it sends a notification to the trained model selection unit 30207 indicating that processing is complete.

[0109] The analysis data holding unit 30206 receives input image analysis data from the analysis data generation unit 30205 and updates the input image group analysis data, which is generated using frames prior to the current frame and held by the analysis data holding unit 30206, by adding the input image analysis data. The method for updating the input image group analysis data is the same as in Example 1, so the explanation is omitted.

[0110] The analysis data storage unit 30206 transmits the updated input image group analysis data to the trained model selection unit 30207.

[0111] Furthermore, an upper limit may be set on the number of images to be used as input image group analysis data. In that case, it is necessary to separately maintain input image analysis data for each frame used to calculate the input image group analysis data up to the upper limit of the number of images, and the oldest input image analysis data is deleted and the latest input image analysis data is added to the set of retained data. The input image group analysis data is calculated each time by adding up the input image analysis data for each frame up to the upper limit of the number of images, according to the conditions.

[0112] The trained model selection unit 30207 selects a trained model to send to the image correction value generation unit 30203 based on the input image group analysis data received from the analysis data holding unit 30206 and the trained models held in the trained model holding unit 30202.

[0113] As a prerequisite, the trained model holder 30202 holds multiple trained models, and the training conditions, such as the images (training images) input to the model training unit 400 in the trained model generation process described later, differ between each trained model. Furthermore, each trained model is associated with training image group analysis data for the images used during training, and the method for generating the training image group analysis data is the same as the method for generating the input image group analysis data described above. Details regarding the training method will be described later.

[0114] The trained model selection unit 30207 searches for and selects the data from each trained image analysis set that is closest to the input image analysis set data, and transmits the trained model associated with the selected trained image analysis set data to the image correction value generation unit 30203. The details of the training model selection method are the same as in Example 1, so the explanation is omitted.

[0115] As explained with reference to Figure 8, in the image brightness adjustment unit 302 of Example 2, the image correction value applied to the input image can be controlled by applying a trained model selected based on the input image group analysis data to the neural network of the image correction value generation unit 30203.

[0116] Figure 9 is a block diagram showing the functional blocks of the model learning unit 400 according to Embodiment 2. The model learning unit 400 comprises a first image input / conversion unit 401, a preprocessing unit 402, an image correction value generation unit 403, an image correction unit 404, a second image input / conversion unit 405, an error calculation unit 406, a trained model output unit 407, an analysis data generation unit 408, and an analysis data holding unit 409. The model learning unit 400 is, for example, a program that runs on a computer, but is not limited to this, and may also operate as a functional block constituting a liquid crystal display device (not shown).

[0117] The first image input / conversion unit 401 is the same as the image input / conversion unit 201 in Embodiment 1, so its description is omitted. The generated input image is transmitted to the preprocessing unit 402, the image correction unit 404, and the analysis data generation unit 408.

[0118] The preprocessing unit 402 generates feature quantities (image feature quantities) based on the input image output from the first image input / conversion unit 401. Specifically, the preprocessing unit 402 generates image feature quantities in the same way as the preprocessing unit 30201 described above. Then, the preprocessing unit 402 outputs the generated image feature quantities to the image correction value generation unit 403.

[0119] The image correction value generation unit 403 consists of a neural network that outputs image correction values ​​based on image features output from the preprocessing unit 402. The neural network of the image correction value generation unit 403 has the same structure as the neural network of the image correction value generation unit 30203 described above. The image correction value generation unit 403 receives the parameters generated by the error calculation unit 406 and reflects them in the neural network. The image correction value generation unit 403 then outputs the image correction values ​​generated by the neural network to the image correction unit 404. In addition, the image correction value generation unit 403 outputs the parameters of the neural network generated by learning as a trained model to the trained model output unit 407.

[0120] The image correction unit 404 generates (calculates) the pixel values ​​of the corrected image by multiplying the image correction values ​​output from the image correction value generation unit 403 by the pixel values ​​of the input image. The calculation method is the same as that of the image correction unit 30204 described above, so the explanation is omitted.

[0121] The second image input / conversion unit 405 performs image conversion on the target image input to the model learning unit 400. The image processing of the second image input / conversion unit 405 is the same as that of the image input / conversion unit 201 in Example 1, so the explanation is omitted. The generated processed target image is sent to the error calculation unit 406.

[0122] The error calculation unit 406 generates parameters for the neural network of the image correction value generation unit 403 based on the difference between the corrected image output from the image correction unit 404 and the processed target image output from the second image input / conversion unit 405. The method of parameter generation by the error calculation unit 406 is the same as that of the error calculation unit 208 in Example 1, so a detailed explanation is omitted.

[0123] The trained model output unit 407 associates the trained model, which is the neural network parameter of the image correction value generation unit 403, with the training image group analysis data held by the analysis data holding unit 409 (described later), and then outputs it externally. For example, the trained model output unit 407 outputs the trained model to the trained model holding unit 30202 in Figure 8 via external memory or the like.

[0124] The analysis data generation unit 408 determines whether the input image (training image) output from the first image input / conversion unit 401 satisfies specific conditions, and outputs the determination result as training image analysis data to the analysis data holding unit 409, which will be described later.

[0125] The method for generating the training image analysis data is the same as that of the analysis data generation unit 30205 in Figure 8 above, so a detailed explanation will be omitted.

[0126] The analysis data holding unit 409 receives the training image analysis data from the analysis data generation unit 408 and updates the training image group analysis data, which is held in the analysis data holding unit 409 and generated from at least frames prior to the current frame, by adding the training image analysis data.

[0127] The specific method for updating the training image set analysis data is the same as that of the analysis data storage unit 30206 in Figure 8, so the explanation will be omitted.

[0128] Furthermore, an upper limit may be set on the number of images to be retained as training image set analysis data. In this case, the method for updating the training image set analysis data is the same as that of the analysis data retention unit 30206 described above, so the explanation will be omitted.

[0129] As explained with reference to Figure 9, the model learning unit 400 of Example 2 can be trained to output an image correction value so that the brightness of the corrected image after image brightness adjustment approaches that of the target image.

[0130] As described in this embodiment, the trained model selection unit 30207 of the image brightness adjustment unit 302 selects a trained model associated with the trained image group analysis data that is closest to the input image group analysis data representing the characteristics of the input content. Therefore, it becomes possible to generate image correction values ​​for brightness adjustment that match the characteristics of the input image.

[0131] The present invention can also be realized by supplying a program that implements one or more of the functions of the above-described embodiments to a system or device via a network or storage medium, and by having one or more processors in the computer of that system or device read and execute the program. It can also be realized by a circuit (e.g., an ASIC) that implements one or more functions.

[0132] [Other embodiments] It should be noted that the present invention is not limited to the embodiments described above, and the components can be modified and implemented in practice without departing from the spirit of the invention. Furthermore, various inventions can be formed by appropriately combining the multiple components disclosed in the above embodiments. For example, some components may be deleted from all the components shown in the embodiments. Moreover, components from different embodiments may be appropriately combined.

[0133] In the above embodiment, at least one of A and B may be A alone, B alone, or A and B.

[0134] Furthermore, the disclosure of this embodiment includes the following configurations and methods. [Configuration 1] LCD panel and An input means for inputting data from the first image group, A backlight module that illuminates the aforementioned liquid crystal panel with light, and whose luminescence brightness can be changed for each of the multiple divided regions, A generation means for generating first image group analysis data from the first image group, A generation means that inputs data generated from the first image included in the first image group into a neural network to generate the luminescence brightness of each divided region of the backlight module, A holding means for holding learned parameters to be set in the neural network used by the generation means, A selection means for determining the learned parameters to be used by the generation means from the holding means based on the first image group analysis data and the learned image group analysis data linked to the learned parameters, A calculation means for calculating a correction value to correct the first image based on the luminescence brightness generated by the generation means, Correction means for correcting the first image to a second image based on the correction value calculated by the calculation means, A light emission control means that controls the light emission brightness of each divided region of the backlight module based on the light emission brightness generated by the generation means, The system includes a display control means for controlling the transmittance of the liquid crystal panel based on the data of the second image, A display device characterized in that the first image group analysis data is data based on the number of images in the first image group that satisfy specific conditions, and the training image group analysis data is data based on images in the training image group that satisfy specific conditions. [Configuration 2] The display device according to configuration 1, characterized in that the data generated from the first image is a feature of the first image. [Configuration 3] The display device according to configuration 1 or 2, characterized in that the selection means selects a trained parameter to which the trained image group analysis data with the closest value to the first image group analysis data is associated. [Structure 4] The system further includes a notification mechanism that prompts the user to retrain the parameters, The display device according to any one of configurations 1 to 3, characterized in that the notification means makes the retraining notification when the similarity between the first image group analysis data and the training image group analysis data falls below a threshold. [Composition 5] The display device according to any one of configurations 1 to 4, characterized in that the generation means generates the first image group analysis data using metadata attached to the first image group. [Composition 6] The system further includes a detection means for detecting when the first image group has been changed to the third image group. When a change is detected in the third image group, the generation means generates third image group analysis data from the third image group. The display device according to any one of configurations 1 to 5, characterized in that the selection means determines the learned parameters to be used by the generation means from the holding means based on the third image group analysis data and the learned image group analysis data linked to the learned parameters. [Composition 7] It further includes a notification mechanism to prompt the user to change parameters, The display device according to any one of configurations 1 to 6, characterized in that the selection means makes a notification to the notification means when changing the selected learned parameters. [Structure 8] The system further includes a reset means for deleting the first image group analysis data, The display device according to configuration 6, characterized in that when the detection means detects a change to the third image group, the reset means deletes the first image group analysis data. [Composition 9] The system further includes a reset means for deleting the first image group analysis data, The display device according to any one of configurations 1 to 7, characterized in that when the user gives an instruction to delete the first image group analysis data, the reset means deletes the first image group analysis data.

Claims

1. LCD panel and An input means for inputting data from the first image group, A backlight module that illuminates the aforementioned liquid crystal panel with light, and whose luminescence brightness can be changed for each of the multiple divided regions, A generation means for generating first image group analysis data from the first image group, A generation means that inputs data generated from the first image included in the first image group into a neural network to generate the luminescence brightness of each divided region of the backlight module, A holding means for holding learned parameters to be set in the neural network used by the generation means, A selection means for determining the learned parameters to be used by the generation means from the holding means based on the first image group analysis data and the learned image group analysis data linked to the learned parameters, A calculation means for calculating a correction value to correct the first image based on the luminescence brightness generated by the generation means, Correction means for correcting the first image to a second image based on the correction value calculated by the calculation means, A light emission control means for controlling the light emission brightness of each divided region of the backlight module based on the light emission brightness generated by the generation means, The system includes a display control means for controlling the transmittance of the liquid crystal panel based on the data of the second image, A display device characterized in that the first image group analysis data is data based on the number of images in the first image group that satisfy specific conditions, and the training image group analysis data is data based on images in the training image group that satisfy specific conditions.

2. The display device according to claim 1, characterized in that the data generated from the first image is a feature of the first image.

3. The display device according to claim 1, characterized in that the selection means selects a trained parameter to which trained image group analysis data with the closest value to the first image group analysis data is associated.

4. The system further includes a notification mechanism that prompts the user to retrain the parameters, The display device according to claim 1, characterized in that the notification means provides the retraining notification when the similarity between the first image group analysis data and the training image group analysis data falls below a threshold.

5. The display device according to claim 1, characterized in that the generation means generates the first image group analysis data using metadata assigned to the first image group as well.

6. The system further includes a detection means for detecting when the first image group has been changed to the third image group. When a change is detected in the third image group, the generation means generates third image group analysis data from the third image group. The display device according to claim 1, characterized in that the selection means determines the learned parameters to be used by the generation means from the holding means based on the third image group analysis data and the learned image group analysis data linked to the learned parameters.

7. It further includes a notification mechanism to prompt the user to change parameters, The display device according to claim 1, characterized in that the selection means provides notification to the notification means when changing the selected learned parameters.

8. The system further includes a reset means for deleting the first image group analysis data, The display device according to claim 6, characterized in that when the detection means detects a change to the third image group, the reset means deletes the first image group analysis data.

9. The system further includes a reset means for deleting the first image group analysis data, The display device according to claim 1, characterized in that when the reset means receives an instruction from the user to delete the first image group analysis data, it deletes the first image group analysis data.

10. A program for causing a computer to function as one of the means described in any one of claims 1 to 9.

11. LCD panel and An input means for inputting data from the first image group, A backlight module that illuminates the aforementioned liquid crystal panel with light, and whose luminescence brightness can be changed for each of the multiple divided regions, A method for controlling a liquid crystal display device having the following characteristics: A generation step of generating first image group analysis data from the first image group, A generation step of inputting data generated from the first image included in the first image group into a neural network to generate the luminescence brightness of each divided region of the backlight module, A holding step for storing the learned parameters to be set in the neural network used in the generation step, A selection step in which the trained parameters to be used in the generation step are determined based on the first image group analysis data and the trained image group analysis data linked to the trained parameters, A calculation step for calculating a correction value to correct the first image based on the luminescence brightness generated in the generation step, A correction step in which the first image is corrected to a second image based on the correction value calculated in the calculation step, A light emission control step that controls the light emission brightness of each divided region of the backlight module based on the light emission brightness generated in the generation step, A display control step that controls the transmittance of the liquid crystal panel based on the data of the second image, A control method having the following characteristics: the first image group analysis data is data based on the number of images in the first image group that satisfy a specific condition, and the learning image group analysis data is data based on the number of images in the learning image group that satisfy a specific condition.

12. The input process involves inputting data from a set of input images, A first generation step involves inputting data generated from the input images included in the aforementioned group of input images into a neural network to generate the luminescence brightness of the backlight module. A second generation step generates a first image based on the data generated from the input image and the luminescence generated by the generation step, A third generation step generates a second image based on data generated from the input image, A calculation step for calculating the difference between the first image and the second image, An update step to update the parameters of the neural network based on the aforementioned difference, A generation step of generating input image group analysis data from the aforementioned input image group, The process involves associating the input image analysis data with the parameters of the neural network, It has, A method for generating a trained model, characterized in that the input image group analysis data is data based on the number of images in the input image group that satisfy specific conditions.

13. The input process involves inputting data from the first image group, A generation step of generating first image group analysis data from the first image group, A generation step of inputting data generated from the first image included in the first image group into a neural network to generate an image correction value for the first image, A holding step for holding the learned parameters to be set in the neural network used in the generation means, A selection step in which the trained parameters to be used in the generation step are determined based on the first image group analysis data and the trained image group analysis data linked to the trained parameters, A correction step in which the first image is corrected to a second image based on the image correction value generated in the generation step, It has, A control method characterized in that the first image group analysis data is data based on the number of images in the first image group that satisfy specific conditions, and the training image group analysis data is data based on the number of images in the training image group that satisfy specific conditions.

14. The input process involves inputting data from a set of input images, A first generation step involves inputting data generated from the input images included in the aforementioned group of input images into a neural network to generate image correction values ​​for the input images. A second generation step generates a first image based on the data generated from the input image and the image correction value generated in the generation step, A third generation step generates a second image based on data generated from the input image, A calculation step for calculating the difference between the first image and the second image, An update step to update the parameters of the neural network based on the aforementioned difference, A generation step of generating input image group analysis data from the aforementioned input image group, The process involves associating the input image analysis data with the parameters of the neural network, It has, A method for generating a trained model, characterized in that the input image group analysis data is data based on the number of images in the input image group that satisfy specific conditions.

15. LCD panel and An input means for inputting a first set of images including the first image, A backlight module capable of changing the luminescence brightness of multiple divided regions, which irradiates light onto the aforementioned liquid crystal panel, A means for holding multiple trained models, A light emission control means that controls the light emission brightness of each of the multiple divided regions of the backlight module using one of the multiple trained models based on the analysis of the first group of images, A display device characterized by having the following features.

16. LCD panel and An input means for inputting a first set of images including the first image, A backlight module that illuminates the aforementioned liquid crystal panel with light, and whose luminescence brightness can be changed for each of the multiple divided regions, A means for holding multiple trained models, A method for controlling a display device having, A control method characterized by having a light emission control step of controlling the light emission brightness of each of the multiple divided regions of the backlight module using one of the multiple trained models based on the analysis of the first group of images.

Images