Picture adjustment method and apparatus

Abstract

The application discloses a picture adjusting method and device, the method comprises the following steps: acquiring a current display picture; performing scene identification processing on the current display picture to obtain a scene identification result corresponding to the current display picture; performing object identification processing on the current display picture to obtain an object identification result corresponding to the current display picture, the object identification result is used for indicating whether a target object exists in the current display picture; performing picture quality adjustment on the current display picture according to the object identification result and the scene identification result, and performing color adjustment on the target object according to the object identification result and the scene identification result. It can be seen that the application can improve the adjusting effect of the picture, and further improve the display effect of the picture.

Description

Technical Field

[0001] This invention relates to the field of display technology, and in particular to a method and apparatus for adjusting screen display. Background Technology

[0002] With the development of display technology, improving the display effect of images has become increasingly important.

[0003] In related technologies, in order to improve the display effect of the image, various adjustments need to be made to the image, such as adjusting the image quality parameters and adjusting the skin tone of the face.

[0004] However, in practice, it was found that the image adjustment was not effective, resulting in poor image display. Summary of the Invention

[0005] The technical problem this invention aims to solve is that poor adjustment of image quality parameters and skin tone in facial features leads to poor image display. Therefore, this invention provides an image adjustment method and apparatus to improve the image adjustment effect, thereby improving the image display effect.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses a screen adjustment method, the method comprising:

[0007] Get the currently displayed screen;

[0008] The current display screen is subjected to scene recognition processing to obtain the scene recognition result corresponding to the current display screen;

[0009] The current display screen is subjected to object recognition processing to obtain the object recognition result corresponding to the current display screen. The object recognition result is used to indicate whether the target object exists in the current display screen.

[0010] The image quality is adjusted based on the object recognition result and the scene recognition result, and the color of the target object is adjusted based on the object recognition result and the scene recognition result.

[0011] As an optional implementation, in the first aspect of the present invention, the scene recognition result includes the scene category corresponding to the currently displayed screen and the initial probability corresponding to each scene category, and the step of adjusting the image quality based on the object recognition result and the scene recognition result includes:

[0012] Based on the object recognition results and the initial probabilities corresponding to each scene category, probability smoothing is performed to obtain the target probability corresponding to each scene category.

[0013] Obtain the image quality parameters corresponding to each of the aforementioned scene categories;

[0014] The target image quality parameters are determined based on the image quality parameters corresponding to each scene category and the target probability corresponding to each scene category.

[0015] Adjust the image quality according to the target image quality parameters.

[0016] As an optional implementation, in the first aspect of the present invention, the step of performing probability smoothing processing based on the object recognition result and the initial probabilities corresponding to each scene category to obtain the target probability corresponding to each scene category includes:

[0017] The initial probabilities corresponding to each scene category are subjected to temporal filtering to obtain the temporal filtered probabilities corresponding to each scene category.

[0018] The priority of each scene category is determined, and the temporal filtering probability of the second scene category is suppressed by the temporal filtering probability of the first scene category in the scene category according to the priority of each scene category, so as to obtain the candidate probability of each scene category. The priority of the first scene category is higher than the priority of the second scene category.

[0019] If the object recognition result indicates that the target object exists in the currently displayed screen, then the candidate probability corresponding to the target scene category associated with the target object is taken as the target probability corresponding to the target scene category, and the candidate probability corresponding to the non-target scene category is subjected to probability change clamping processing to obtain the target probability corresponding to the non-target scene category. The non-target scene category includes categories other than the target scene category in the scene category.

[0020] As an optional implementation, in the first aspect of the present invention, the step of performing probability change clamping processing on the candidate probabilities corresponding to non-target scene categories to obtain the target probabilities corresponding to the non-target scene categories includes:

[0021] Based on the candidate probability corresponding to the non-target scene category and the initial probability corresponding to the non-target scene category, determine the probability change value of the non-target scene category;

[0022] Determine whether the probability change value is greater than the probability change threshold;

[0023] If the probability change value is less than or equal to the probability change threshold, then the candidate probability corresponding to the non-target scene category is taken as the target probability corresponding to the non-target scene category.

[0024] If the probability change value is greater than the probability change threshold, then the target probability corresponding to the non-target scene category is determined based on the initial probability corresponding to the non-target scene category and the probability change threshold. The target probability corresponding to the non-target scene category is positively correlated with the initial probability corresponding to the non-target scene category and the probability change threshold.

[0025] As an optional implementation, in the first aspect of the present invention, the suppression processing of the temporal filtering probability corresponding to the second scene category in the scene category using the temporal filtering probability corresponding to the first scene category in the scene category includes:

[0026] Determine the maximum prediction probability, which is the maximum value among the temporal filtered probabilities corresponding to the first scene category;

[0027] The inhibition factor is determined by multiplying the maximum predicted probability by the preset inhibition coefficient.

[0028] The candidate probability corresponding to the second scene category is determined based on the difference between the temporal filtering probability and the suppression probability corresponding to the second scene category. The suppression probability is obtained by multiplying the temporal filtering probability corresponding to the second scene category with the suppression factor.

[0029] As an optional implementation, in the first aspect of the present invention, the scene recognition result includes the scene category corresponding to the current display screen and the initial probability corresponding to each scene category, and the object recognition result is further used to represent the current position information of the target object in the current display screen. The step of adjusting the color of the target object based on the object recognition result and the scene recognition result includes:

[0030] Determine the target scene category associated with the target object, and obtain the initial probability corresponding to the target scene category based on the initial probability corresponding to each scene category;

[0031] The target weight parameter for color adjustment of the target object is determined based on the initial probability corresponding to the target scene category.

[0032] Based on the current location information, determine the target area occupied by the target object in the current display screen, and statistically analyze the HSV color information of the target area;

[0033] The color of the target object is adjusted according to the target weight parameters and the HSV color information.

[0034] As an optional implementation, in the first aspect of the present invention, determining the target weight parameter for color adjustment of the target object based on the initial probability corresponding to the target scene category includes:

[0035] A scene reference weight is determined based on the initial probability corresponding to the target scene category, and the scene reference weight is positively correlated with the initial probability corresponding to the target scene category.

[0036] A color reference weight is determined based on the object recognition result. The first color reference weight is greater than the second color reference weight. The first color reference weight is the weight corresponding to the presence of the target object in the current display screen, and the second color reference weight is the weight corresponding to the absence of the target object in the current display screen.

[0037] The target weight parameter is determined based on the scene reference weight and the color reference weight, and the target weight parameter is positively correlated with the scene reference weight and the color reference weight, respectively.

[0038] As an optional implementation, in the first aspect of the present invention, the currently displayed screen is a video frame of a video stream, and the step of determining the target area occupied by the target object in the currently displayed screen based on the current position information includes:

[0039] Obtain N frames of reference display images, wherein the reference display images include images that were displayed before the current display image and contain the target object, and N is a positive integer greater than or equal to 1;

[0040] For each frame of the reference display image, position smoothing processing is performed to obtain at least one candidate position information;

[0041] The target region is determined based on the at least one candidate location information;

[0042] The position smoothing process includes:

[0043] The IOU value of the target object is determined based on the current area occupied by the target object in the current display screen and the reference area occupied by the target object in the reference display screen.

[0044] If the IOU value is less than the IOU threshold, the candidate position information is determined based on the current position information and the average value of the reference position information of the target object in the reference display screen.

[0045] If the IOU value is greater than or equal to the IOU threshold, the candidate position information is determined based on the weighted average of the current position information and the reference position information of the target object in the reference display screen, wherein the reference display screen with more video frames between it and the current display screen has a smaller weight.

[0046] A second aspect of the present invention discloses a screen adjustment device, the device comprising:

[0047] The acquisition module is used to acquire the currently displayed screen.

[0048] The scene recognition module is used to perform scene recognition processing on the current display screen to obtain the scene recognition result corresponding to the current display screen;

[0049] An object recognition module is used to perform object recognition processing on the current display screen to obtain an object recognition result corresponding to the current display screen. The object recognition result is used to indicate whether a target object exists in the current display screen.

[0050] The image adjustment module is used to adjust the image quality based on the object recognition result and the scene recognition result, and to adjust the color of the target object based on the object recognition result and the scene recognition result.

[0051] As an optional implementation, in a second aspect of the present invention, the scene recognition result includes the scene category corresponding to the currently displayed screen and the initial probability corresponding to each scene category, and the screen adjustment module includes:

[0052] A probability smoothing unit is used to perform probability smoothing processing based on the object recognition result and the initial probability corresponding to each scene category to obtain the target probability corresponding to each scene category.

[0053] The parameter acquisition unit is used to acquire the image quality parameters corresponding to each of the scene categories;

[0054] The first adjustment unit is used to determine the target image quality parameters based on the image quality parameters corresponding to each scene category and the target probability corresponding to each scene category; and to adjust the image quality of the image based on the target image quality parameters.

[0055] As an optional implementation, in a second aspect of the invention, the probability smoothing unit includes:

[0056] The temporal filtering and smoothing subunit is used to perform temporal filtering on the initial probabilities corresponding to each of the scene categories to obtain the temporal filtered probabilities corresponding to each of the scene categories.

[0057] The priority suppression subunit is used to determine the priority corresponding to each scene category, and suppress the temporal filtering probability corresponding to the second scene category according to the priority corresponding to each scene category through the temporal filtering probability corresponding to the first scene category in the scene category, so as to obtain the candidate probability corresponding to each scene category, wherein the priority corresponding to the first scene category is higher than the priority corresponding to the second scene category.

[0058] The change clamping subunit is used to, if the object recognition result indicates that the target object exists in the current display screen, take the candidate probability corresponding to the target scene category associated with the target object as the target probability corresponding to the target scene category, and perform probability change clamping processing on the candidate probabilities corresponding to non-target scene categories to obtain the target probability corresponding to the non-target scene category. The non-target scene category includes categories other than the target scene category in the scene category.

[0059] As an optional implementation, in a second aspect of the present invention, the change clamping subunit is used to determine the probability change value of the non-target scene category based on the candidate probability corresponding to the non-target scene category and the initial probability corresponding to the non-target scene category.

[0060] Determine whether the probability change value is greater than the probability change threshold;

[0061] If the probability change value is less than or equal to the probability change threshold, then the candidate probability corresponding to the non-target scene category is taken as the target probability corresponding to the non-target scene category.

[0062] If the probability change value is greater than the probability change threshold, then the target probability corresponding to the non-target scene category is determined based on the initial probability corresponding to the non-target scene category and the probability change threshold. The target probability corresponding to the non-target scene category is positively correlated with the initial probability corresponding to the non-target scene category and the probability change threshold.

[0063] As an optional implementation, in a second aspect of the invention, a priority suppression subunit is used to determine the maximum prediction probability, which is the maximum value among the temporal filtering probabilities corresponding to the first scene category;

[0064] The inhibition factor is determined by multiplying the maximum predicted probability by the preset inhibition coefficient.

[0065] The candidate probability corresponding to the second scene category is determined based on the difference between the temporal filtering probability and the suppression probability corresponding to the second scene category. The suppression probability is obtained by multiplying the temporal filtering probability corresponding to the second scene category with the suppression factor.

[0066] As an optional implementation, in a second aspect of the present invention, the scene recognition result includes the scene category corresponding to the current display screen and the initial probability corresponding to each scene category; the object recognition result is further used to represent the current position information of the target object in the current display screen; and the screen adjustment module further includes:

[0067] A probability acquisition unit is used to determine the target scene category associated with the target object, and to acquire the initial probability corresponding to the target scene category according to the initial probability corresponding to each scene category;

[0068] The weight determination unit is used to determine the target weight parameters for color adjustment of the target object based on the initial probability corresponding to the target scene category.

[0069] The HSV statistics unit is used to determine the target area occupied by the target object in the current display screen based on the current location information, and to count the HSV color information of the target area;

[0070] The second adjustment unit is used to adjust the color of the target object according to the target weight parameter and the HSV color information.

[0071] As an optional implementation, in a second aspect of the present invention, the weight determination unit is used to determine a scene reference weight based on the initial probability corresponding to the target scene category, wherein the scene reference weight is positively correlated with the initial probability corresponding to the target scene category;

[0072] A color reference weight is determined based on the object recognition result. The first color reference weight is greater than the second color reference weight. The first color reference weight is the weight corresponding to the presence of the target object in the current display screen, and the second color reference weight is the weight corresponding to the absence of the target object in the current display screen.

[0073] The target weight parameter is determined based on the scene reference weight and the color reference weight, and the target weight parameter is positively correlated with the scene reference weight and the color reference weight, respectively.

[0074] As an optional implementation, in the second aspect of the present invention, the current display screen is one video frame of a video stream, and the HSV statistics unit is used to obtain N reference display screens, the reference display screens including screens displayed before the current display screen that contain the target object, and N is a positive integer greater than or equal to 1;

[0075] For each frame of the reference display image, position smoothing processing is performed to obtain at least one candidate position information;

[0076] The target region is determined based on the at least one candidate location information;

[0077] The position smoothing process includes:

[0078] The IOU value of the target object is determined based on the current area occupied by the target object in the current display screen and the reference area occupied by the target object in the reference display screen.

[0079] If the IOU value is less than the IOU threshold, the candidate position information is determined based on the current position information and the average value of the reference position information of the target object in the reference display screen.

[0080] If the IOU value is greater than or equal to the IOU threshold, the candidate position information is determined based on the weighted average of the current position information and the reference position information of the target object in the reference display screen, wherein the reference display screen with more video frames separated from the current display screen has a smaller weight.

[0081] A third aspect of the present invention discloses another screen adjustment device, the device comprising:

[0082] Memory containing executable program code;

[0083] A processor coupled to the memory;

[0084] The processor calls the executable program code stored in the memory to execute the screen adjustment method disclosed in the first aspect of the present invention.

[0085] The fourth aspect of the present invention discloses a computer-storable medium storing computer instructions, which, when invoked, are used to execute the screen adjustment method disclosed in the first aspect of the present invention.

[0086] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0087] In this embodiment of the invention, the current display screen is acquired; scene recognition processing is performed on the current display screen to obtain a scene recognition result corresponding to the current display screen; object recognition processing is performed on the current display screen to obtain an object recognition result corresponding to the current display screen, the object recognition result indicating whether a target object exists in the current display screen; the image quality is adjusted based on the object recognition result and the scene recognition result, and the color of the target object is adjusted based on the object recognition result and the scene recognition result; that is, this embodiment adjusts the image quality and the target object's color based on the scene recognition result and the object recognition result. Therefore, implementing this invention can improve the image quality adjustment result and simultaneously improve the target object's color adjustment result, thus improving the image adjustment effect and consequently the display effect. Attached Figure Description

[0088] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0089] Figure 1 This is a flowchart illustrating a screen adjustment method disclosed in an embodiment of the present invention;

[0090] Figure 2 This is a flowchart illustrating another screen adjustment method disclosed in an embodiment of the present invention;

[0091] Figure 3 This is a schematic diagram of the structure of a screen adjustment device disclosed in an embodiment of the present invention;

[0092] Figure 4 This is a schematic diagram of the structure of another screen adjustment device disclosed in an embodiment of the present invention. Detailed Implementation

[0093] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0094] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0095] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0096] This invention discloses a screen adjustment method and apparatus, which can improve the screen adjustment effect and thus improve the screen display effect. Detailed descriptions follow.

[0097] Example 1

[0098] Please see Figure 1 , Figure 1 This is a flowchart illustrating a screen adjustment method disclosed in an embodiment of the present invention. Figure 1 The described image adjustment method can be applied to an image adjustment device. This image adjustment device can be a device with image display capabilities, such as a television or mobile terminal, or it can be a unit with computing power within such a device, such as a processor. This embodiment does not limit the scope. Figure 1 As shown, the screen adjustment method may include steps 110 to 140.

[0099] Step 110: Obtain the currently displayed screen.

[0100] The currently displayed frame refers to one of the frames shown. Specifically, the currently displayed frame can be a static image; in addition, the currently displayed frame can also be a frame from a video stream, meaning that the currently displayed frame can be dynamically changing.

[0101] Step 120: Perform scene recognition processing on the current display screen to obtain the scene recognition result corresponding to the current display screen.

[0102] In this embodiment, scene recognition processing can be the process of identifying the scene category of the currently displayed image. Optionally, the scene recognition module can process the currently displayed image to obtain the scene recognition result corresponding to the currently displayed image.

[0103] In one possible implementation, the scene recognition module processes the currently displayed image to obtain the scene recognition result corresponding to the currently displayed image, which may include:

[0104] Extract the current screen features of the currently displayed screen;

[0105] The current screen features are input into the trained scene recognition model, which is used to perform scene recognition processing on the current screen features to obtain the scene recognition result corresponding to the current display screen.

[0106] Optionally, the scene recognition model can be a neural network model, which can be trained by combining image features and scene category labels to obtain a trained scene recognition model.

[0107] Step 130: Perform object recognition processing on the current display screen to obtain the object recognition result corresponding to the current display screen.

[0108] The object recognition result is used to indicate whether a target object exists in the currently displayed screen. In this embodiment, the object recognition process includes identifying whether a target object exists in the current screen. Optionally, the object recognition module can perform object recognition processing on the current displayed screen. Specifically, the target object includes, but is not limited to, a face, flowers, blue sky, etc., and the target object can be set as needed, without limitation here. Specifically, taking a face as an example, the object recognition process is a face recognition process, and the object recognition module can be a face recognition module.

[0109] Step 140: Adjust the image quality based on the object recognition result and the scene recognition result, and adjust the color of the target object based on the object recognition result and the scene recognition result.

[0110] The image quality adjustment can be applied to the entire image. Optionally, image quality adjustment includes, but is not limited to, noise reduction, sharpening, color enhancement, and contrast enhancement. In this embodiment, if the target object is a human face, it can be an adjustment of the skin tone, etc.

[0111] Specifically, in these related technologies, image quality parameters are generally adjusted solely based on scene recognition results. Skin tone adjustment for faces typically relies on face detection results and HSV values ​​within the face region for skin color protection. Therefore, these technologies do not utilize other recognition results for adjustment, leading to suboptimal image quality.

[0112] The technical solution of this embodiment involves: acquiring the current display screen; performing scene recognition processing on the current display screen to obtain a scene recognition result corresponding to the current display screen; performing object recognition processing on the current display screen to obtain an object recognition result corresponding to the current display screen, wherein the object recognition result indicates whether a target object exists in the current display screen; adjusting the image quality based on the object recognition result and the scene recognition result; and adjusting the color of the target object based on the object recognition result and the scene recognition result. In other words, this embodiment improves the image quality and the target object color by adjusting the image quality and the target object color through the scene recognition and object recognition results. This improves both the image quality and the target object color, thereby enhancing the image adjustment effect and ultimately improving the display effect.

[0113] It is understandable that steps 120 and 130 can be performed asynchronously or synchronously, and no restriction is made here.

[0114] The following embodiments, based on any of the above embodiments, respectively illustrate how to adjust the image quality based on the object recognition result and the scene recognition result, and how to adjust the color of the target object based on the object recognition result and the scene recognition result.

[0115] First, regarding how to adjust the image quality based on the object recognition results and the scene recognition results.

[0116] In one possible implementation, the scene recognition result includes the scene category corresponding to the currently displayed screen and an initial probability corresponding to each scene category, wherein the initial probability can be understood as the predicted probability. The step of adjusting the image quality based on the object recognition result and the scene recognition result includes:

[0117] Based on the object recognition results and the initial probabilities corresponding to each scene category, probability smoothing is performed to obtain the target probability corresponding to each scene category.

[0118] Obtain the image quality parameters corresponding to each of the aforementioned scene categories;

[0119] The target image quality parameters are determined based on the image quality parameters corresponding to each scene category and the target probability corresponding to each scene category.

[0120] Adjust the image quality according to the target image quality parameters.

[0121] The scene categories include, but are not limited to, portraits, blue skies, grasslands, and flowers, which are determined based on the scene recognition results and are not limited here.

[0122] In this embodiment, each scene category can be pre-tuned with a set of optimal image quality parameters. These sets of image quality parameters are weighted and fused together based on the smoothed target probability to form the target image quality parameters most suitable for the current scene. The image quality is then adjusted according to the target image quality parameters to achieve the best display effect.

[0123] In one possible implementation, probability smoothing is performed based on the object recognition result and the initial probabilities corresponding to each scene category to obtain the target probability corresponding to each scene category, including:

[0124] The initial probabilities corresponding to each scene category are subjected to temporal filtering to obtain the temporal filtered probabilities corresponding to each scene category.

[0125] The priority of each scene category is determined, and the temporal filtering probability of the second scene category is suppressed by the temporal filtering probability of the first scene category in the scene category according to the priority of each scene category, so as to obtain the candidate probability of each scene category. The priority of the first scene category is higher than the priority of the second scene category.

[0126] If the object recognition result indicates that the target object exists in the currently displayed screen, then the candidate probability corresponding to the target scene category associated with the target object is taken as the target probability corresponding to the target scene category, and the candidate probability corresponding to the non-target scene category is subjected to probability change clamping processing to obtain the target probability corresponding to the non-target scene category. The non-target scene category includes categories other than the target scene category in the scene category.

[0127] For example, if the target object is a face, then the target scene category is a face scene.

[0128] In this embodiment, filtering the initial probabilities in the time domain prevents noise from affecting the results, making the predicted probabilities smoother. The filtering method can be an arithmetic mean filter or a weighted filter applied to the initial probabilities of the past 5 or 10 probability iterations. The filtering process can be represented as follows:

[0129] PRf = F(PRraw, Ht)

[0130] Where F(PRraw, Ht) represents applying a low-pass filter with kernel Ht to the initial probability PRraw.

[0131] In this embodiment, the probability of error varies among the scene categories obtained from scene recognition processing. For example, the probability of error is relatively low in categories such as portraits, blue skies, and grasslands, while the probability of error is much higher in scenes like flowers. This is because these scene categories in the real world are more complex and have more features. Therefore, a priority mechanism is used here, designating categories such as portraits, blue skies, and grasslands as the first scene category with high priority, while flowers are designated as the second scene category with low priority. This setting leads to the following special processing method: the prediction results of high-priority categories will suppress the prediction results of low-priority categories, preventing the low-priority categories from interfering with the output results.

[0132] In this embodiment, if the predicted probability of the scene category changes too rapidly, it will cause flickering in the display. Therefore, the change in probability needs to be kept within a reasonable range. Furthermore, when the object recognition result indicates that the target object exists in the current display screen, it proves that the target object is present in the screen, and the probability of the target scene category associated with the target object must not decrease.

[0133] It is understood that this embodiment improves the accuracy of determining the target probability corresponding to each scene category by performing temporal filtering on the initial probability corresponding to each scene category, and by using the temporal filtering probability corresponding to the high-priority scene type to filter the temporal filtering probability corresponding to the low-priority scene type, and by keeping the change value of the clamping probability within a reasonable range. This further improves the accuracy of image quality adjustment.

[0134] The following examples further illustrate how to suppress probabilities by prioritization.

[0135] In one possible implementation, the temporal filtering probability corresponding to the second scene category in the scene category is suppressed using the temporal filtering probability corresponding to the first scene category in the scene category, including:

[0136] Determine the maximum prediction probability, which is the maximum value among the temporal filtered probabilities corresponding to the first scene category;

[0137] The inhibition factor is determined by multiplying the maximum predicted probability by the preset inhibition coefficient.

[0138] The candidate probability corresponding to the second scene category is determined based on the difference between the temporal filtering probability and the suppression probability corresponding to the second scene category. The suppression probability is obtained by multiplying the temporal filtering probability corresponding to the second scene category with the suppression factor.

[0139] For example, the scene categories include face scenes, blue sky scenes, grass scenes, and flower scenes, with face scenes, blue sky scenes, and grass scenes having higher priority than flower scenes.

[0140] First, calculate the maximum predicted probability for the current high-priority category:

[0141] PRmax=max(PRf_p, PRf_b, PRf_g)

[0142] Where PRmax is the maximum predicted probability of the high-priority category, and PRf_p, PRf_b, and PRf_g are the predicted probabilities of the portrait, blue sky, and grass scenes after temporal filtering, respectively.

[0143] Then calculate the inhibition factor:

[0144] Flim = Alim * PRmax

[0145] The Alim value can be preset, for example, set to 0.01.

[0146] After calculating the suppression factor Flim, the prediction results for low-priority classes can be restricted, which can be expressed as:

[0147] PRf_low=max(PRf_low-Flim*PRmax, 0)

[0148] Where PRf_low represents the target probability of each low-priority category.

[0149] It should be noted that in this embodiment, the priority level can be set to two levels or more; no limitation is made here. Specifically, if the priority level is set to three or more, it is necessary to differentiate between each level for probability clamping of change values.

[0150] For example, if the design uses three priorities: high, medium, and low, it would be as follows:

[0151] The probability of high-priority categories remains unchanged: PRhigh.

[0152] The probability of medium-priority categories will be suppressed by higher-priority categories:

[0153] PRmid = PRmid - Alim * PRhigh.

[0154] The probability of low-priority categories is suppressed by high-priority categories:

[0155] PRlow=PRlow-Alim*PRhigh-Blim*PRmid.

[0156] The Blim value can be preset, for example, set to 0.01.

[0157] The following examples further illustrate how to perform probability change clamping.

[0158] In one possible implementation, the candidate probabilities corresponding to non-target scene categories are subjected to probability change clamping processing to obtain the target probability corresponding to the non-target scene category, including:

[0159] Based on the candidate probability corresponding to the non-target scene category and the initial probability corresponding to the non-target scene category, determine the probability change value of the non-target scene category;

[0160] Determine whether the probability change value is greater than the probability change threshold;

[0161] If the probability change value is less than or equal to the probability change threshold, then the candidate probability corresponding to the non-target scene category is taken as the target probability corresponding to the non-target scene category.

[0162] If the probability change value is greater than the probability change threshold, then the target probability corresponding to the non-target scene category is determined based on the initial probability corresponding to the non-target scene category and the probability change threshold. The target probability corresponding to the non-target scene category is positively correlated with the initial probability corresponding to the non-target scene category and the probability change threshold.

[0163] For example, suppose the initial probability of the input is PRain, the result of time-domain filtering is PRf, the result of priority clamping is PRL, and the result of change value clamping is PRres.

[0164] PRf = Time-domain filtering (PRin);

[0165] PRl = Priority Restriction (PRf).

[0166] Assuming the change value needs to be clamped within range A, we determine if the difference between PRL and PRin is greater than A. If the difference between PRL and PRin is less than or equal to A, then PRres = PRRL, meaning the result of priority clamping is used as the target probability. If the difference between PRL and PRin is greater than A, then PRres = PRin + sign * A. Here, sign can be any value between [-1, 1]. Optionally, if PRL - PRin ≥ A, then sign is 1; if PRL - PRin < A, then sign is -1.

[0167] It is understood that in the above embodiments, the image quality is adjusted by the object recognition result and the scene recognition result. That is to say, the image quality adjustment of the whole frame takes into account the object recognition result and the scene recognition result, thus improving the image quality adjustment effect.

[0168] The following embodiments, based on any of the above embodiments, illustrate how to adjust the color of the target object according to the object recognition result and the scene recognition result.

[0169] In one possible implementation, the scene recognition result includes the scene category corresponding to the current display screen and the initial probability corresponding to each scene category. The object recognition result is further used to represent the current position information of the target object in the current display screen. The step of adjusting the color of the target object based on the object recognition result and the scene recognition result includes:

[0170] Determine the target scene category associated with the target object, and obtain the initial probability corresponding to the target scene category based on the initial probability corresponding to each scene category;

[0171] The target weight parameter for color adjustment of the target object is determined based on the initial probability corresponding to the target scene category.

[0172] Based on the current location information, determine the target area occupied by the target object in the current display screen, and statistically analyze the HSV color information of the target area;

[0173] The color of the target object is adjusted according to the target weight parameters and the HSV color information.

[0174] In this embodiment, the color of the target object is adjusted according to the target weight parameter and HSV color information so that the target object does not change color.

[0175] It should be noted that in this embodiment, when a target object is detected in the current display screen, the current position information of the target object in the current display screen will be output. At this time, the presence of the target object in the current display screen can be confirmed by the current position information. Optionally, if the current position information is not output, it is assumed that the target object does not exist in the current display screen.

[0176] In one possible implementation, determining the target weight parameter for color adjustment of the target object based on the initial probability corresponding to the target scene category includes:

[0177] A scene reference weight is determined based on the initial probability corresponding to the target scene category, and the scene reference weight is positively correlated with the initial probability corresponding to the target scene category.

[0178] A color reference weight is determined based on the object recognition result. The first color reference weight is greater than the second color reference weight. The first color reference weight is the weight corresponding to the presence of the target object in the current display screen, and the second color reference weight is the weight corresponding to the absence of the target object in the current display screen.

[0179] The target weight parameter is determined based on the scene reference weight and the color reference weight, and the target weight parameter is positively correlated with the scene reference weight and the color reference weight, respectively.

[0180] In this embodiment, for example, the target weight parameter is denoted as skin_gain, with a value range of [0, 255]. The result of skin_gain is determined by the scene recognition result and the object recognition result in scene detection. It can be expressed as:

[0181] skin_gain=scene_gain+face_gain.

[0182] The weight of scene_gain in skin_gain is scene_gain_max, the weight of face_gain in skin_gain is face_gain_max, and scene_gain_max plus face_gain_max equals 255.

[0183] scene_gain can be obtained by predicting the probability PRF_p of the portrait scene:

[0184] scene_gain=PRf_p*(scene_gain_max / 100)

[0185] The face_gain can be determined by the object recognition result. If a target object is detected in the image, face_gain equals face_gain_max; otherwise, face_gain equals 0.

[0186] In this embodiment, the target weight parameter is determined by combining the scene recognition result and the object recognition result, and then the target weight parameter is used to adjust the color of the target object, which can improve the accuracy of the color adjustment of the target object.

[0187] In one possible implementation, the currently displayed frame is a video frame from a video stream, and determining the target area occupied by the target object in the currently displayed frame based on the current position information includes:

[0188] Obtain N frames of reference display images, wherein the reference display images include images that were displayed before the current display image and contain the target object, and N is a positive integer greater than or equal to 1;

[0189] For each frame of the reference display image, position smoothing processing is performed to obtain at least one candidate position information;

[0190] The target region is determined based on the at least one candidate location information;

[0191] The position smoothing process includes:

[0192] The IOU value of the target object is determined based on the current area occupied by the target object in the current display screen and the reference area occupied by the target object in the reference display screen.

[0193] If the IOU value is less than the IOU threshold, the candidate position information is determined based on the current position information and the average value of the reference position information of the target object in the reference display screen.

[0194] If the IOU value is greater than or equal to the IOU threshold, the candidate position information is determined based on the weighted average of the current position information and the reference position information of the target object in the reference display screen, wherein the reference display screen with more video frames separated from the current display screen has a smaller weight.

[0195] In this embodiment, N can be set as needed and is not limited here, for example, 5, 10, etc. Optionally, the N frames of reference display images can be N consecutive reference display images adjacent to the current display image, and is not limited here. Optionally, the current position information includes, but is not limited to, the center coordinates, width, and height of the target object, etc., and is not limited here.

[0196] In this embodiment, each reference display screen corresponds to a candidate location information. The information in the candidate location information is averaged and summed to obtain the target location information, and then the target area is determined through the target location information.

[0197] For example, define the coordinates of 5 target objects as Coor1~5, and the current frame is Coor1.

[0198] Four IOU values, IOU2 to IOU5, are obtained by calculating Coor1 with other coordinates.

[0199] If max(IOU2~5) is small, then the new Coor_Output is obtained by average filtering:

[0200] Coor_Output=mean(Coor1~5).

[0201] If max(IOU2~5) is large, then weighted filtering should be used.

[0202] Coor_Output=(1*Coor1+IOU2*Coor2+IOU3*Coor3+IOU4*Coor4+

[0203] IOU5*Coor5) / (1+IOU2+IOU3+IOU4+IOU5).

[0204] In this embodiment, the center coordinates, width, and height information in the current location information can all be calculated using the methods described above.

[0205] The technical solution of this embodiment determines the target area occupied by the target object by using the current position information of the currently displayed screen and the reference position information of the target object on the reference display screen, which can improve the accuracy of target area determination and further improve the accuracy of screen adjustment.

[0206] Optionally, adjusting the color of the target object based on the target weight parameter and the HSV color information may include:

[0207] Determine the color judgment threshold of the target object based on the HSV color information;

[0208] Obtain the color detection results of each pixel in the target area. The color detection results are used to indicate the probability that the pixel is the target object. The color detection results are related to the color judgment threshold.

[0209] Based on the target weight parameters and the color detection results of each pixel, the target color detection result of each pixel is determined;

[0210] The color of each pixel is adjusted based on the target color detection results corresponding to each pixel.

[0211] For example, let the color detection result of a pixel be PRdet, and the target weight parameter be skin_gain, PRdet = skin_gain * PRdet / 255 (skin_gain ∈ (0, 255)).

[0212] In this embodiment, adjusting the color detection threshold of pixels based on target object parameters and HSV color information can improve the accuracy of color detection of the target object.

[0213] Specifically, in the above embodiments, the target object can be a human face, flowers, blue sky, etc. The following embodiments use a human face as an example to illustrate this solution.

[0214] Please see Figure 2 , Figure 2 This is a flowchart illustrating another screen adjustment method disclosed in an embodiment of the present invention.

[0215] I. Feature Analysis of Neural Network Models

[0216] In this invention, feature analysis includes scene classification and face detection. Scene classification determines which scene the current image belongs to and outputs the similarity of each scene (i.e., prediction result, prediction probability, and initial probability). The default scene classifications are portrait, blue sky, grass, and flowers. Categories can be modified as needed. Face detection detects whether there are faces in the image; if faces are present, it outputs the coordinates of each face.

[0217] II. Processing of Prediction Results

[0218] The prediction result processing section processes the output of the neural network model and simultaneously outputs configuration information to the image enhancement section.

[0219] 2.1 Probability Smoothing

[0220] Scene classification models can determine which scene an image belongs to and output the similarity of each scene. However, in real-world video content, due to the large variety of objects and fast camera movement, the predicted probabilities of neural network models are sometimes unstable. Without special processing, this can easily cause screen flickering and color cast. Therefore, a probability smoothing module is needed to stabilize the probabilities. Specifically, smoothing involves time-domain filtering, priority constraints, and value clamping.

[0221] 2.1.1 Time-domain filtering

[0222] Filtering the predicted probabilities in the time domain prevents noise from affecting the results and makes the predicted probabilities smoother. Filtering methods can include arithmetic mean filtering or weighted filtering based on the predicted probabilities of the past 5 or 10 predictions. The filtering process can be expressed as:

[0223] PRf = F(PRraw, Ht)

[0224] Where F(PRraw,Ht) represents the application of a low-pass filter with kernel Ht to the initial predicted probability PRraw.

[0225] 2.1.2 Priority Suppression Limitation

[0226] The probability of a neural network model making an error varies across different scenarios. For example, the probability of error is relatively low in categories like portraits, blue skies, and grasslands, while it is much higher in scenarios like flowers. This is because these scenarios in the real world are more complex, have more features, and the training set cannot cover all situations. Errors in these scenarios can easily interfere with the prediction results of the entire model. Therefore, a priority mechanism is used here, setting categories like portraits, blue skies, and grasslands as high priority and flowers as low priority. This setting results in the following special processing method: the prediction results of high-priority categories will suppress the prediction results of low-priority categories, preventing the low-priority categories from interfering with the output results. The specific process can be represented as follows:

[0227] First, calculate the maximum predicted probability for the current high-priority category:

[0228] PRmax=max(PRf_p,PRf_b,PRf_g)

[0229] Where PRmax is the maximum predicted probability of the high-priority category, and PRf_p, PRf_b, and PRf_g are the predicted probabilities of the portrait, blue sky, and grass scenes after temporal filtering, respectively.

[0230] Then calculate the inhibition factor:

[0231] Flim = Alim * PRmax

[0232] The Alim value can be 0.01.

[0233] After calculating the suppression factor Flim, the prediction results for low-priority classes can be restricted, which can be expressed as:

[0234] PRf_low=max(PRf_low-Flim*PRmax,0)

[0235] Where PRf_low represents the prediction results for each low-priority category.

[0236] 2.1.3 Probability Change Value Clamping

[0237] If the predicted probability changes too rapidly, it will cause flickering in the display. Therefore, the change in probability needs to be kept within a reasonable range. Furthermore, when the face detection module detects a face, it indicates that a person is present in the image, and the predicted probability for portrait scenes should not be reduced.

[0238] 2.2 Weighted fusion of image quality parameters

[0239] Each scene category has a pre-calibrated set of optimal image quality parameters. These image quality parameters are then weighted and merged based on the smoothed probability to form the most suitable image quality parameters for the current scene. These parameters are then configured for subsequent image quality enhancement modules to achieve the best display effect.

[0240] 2.3 Skin Color Protection Weight Calculation

[0241] The skin color protection module selects the range of skin tones by setting a threshold range in the HSV color space. This method is simple, efficient, and has low side effects. However, this method will classify all colors within the skin color range as skin tones. Therefore, the prediction results of a neural network model are needed to assist in the determination. Skin color detection is not performed on non-human images. Experiments showed that simply using scene classification or face detection to determine whether there are human figures in an image is prone to errors; therefore, this invention requires the simultaneous use of both models for determination.

[0242] Let the skin color protection weight be skin_gain, with a value range of [0, 255]. The result of Skin_gain is determined by the portrait scene prediction result and the face detection result in scene detection. It can be expressed as:

[0243] skin_gain=scene_gain+face_gain

[0244] The weight of scene_gain in skin_gain is scene_gain_max, the weight of face_gain in skin_gain is face_gain_max, and scene_gain_max plus face_gain_max equals 255.

[0245] scene_gain can be obtained from the portrait scene prediction result PRF_p:

[0246] scene_gain=PRf_p*(scene_gain_max / 100)

[0247] The face_gain can be determined by the face detection results. If a face is detected in the image, face_gain equals face_gain_max; otherwise, face_gain equals 0.

[0248] The final result of skin_gain indicates the probability of a human figure appearing in the image, which can be used to guide the intensity settings of the skin color detection module.

[0249] 2.4 Face coordinate smoothing

[0250] This module improves the stability of face detection results in video by filtering and smoothing the results in consecutive video frames, while ensuring the accuracy of face location. Faces in consecutive video frames are correlated using Intersection over Union (IoU) calculation. Then, the center coordinates, width, height, and top-left corner coordinates of the correlated faces in each consecutive video frame are filtered and smoothed to obtain the face location coordinates of the current frame. Here, the number of consecutive frames is set to 5 or 10, and the filtering weights used for smoothing are related to the IoU value. When the IoU value is large, average filtering is used; when the IoU value is small, larger weights are applied to frames that are temporally adjacent, and smaller weights are applied to frames that are temporally distant. The IoU value is calculated as the area of ​​the intersection of two face bounding boxes divided by the area of ​​their combined shape.

[0251] 2.5 HSV value statistics within the face frame

[0252] This module is used to calculate the average values ​​of chroma (H), saturation (S), and luminance (V) within a face frame. It assists the skin tone detection module in selecting the skin tone range within the HSV color space, making skin tone detection more accurate.

[0253] III. Image Quality Enhancement

[0254] The image quality enhancement section includes an image quality enhancement module and a skin color protection module. Image quality enhancement can be achieved through sub-modules such as noise reduction, sharpening, color enhancement, and contrast enhancement. Image quality parameters can adjust the intensity and operation of each sub-module. The optimal parameters differ for each scene. In this invention, the image quality parameters for each scene are automatically generated based on the scene classification results to achieve the best image quality effect. The skin color protection module is used to detect skin color and prevent color changes. This module utilizes the skin color protection weight `skin_gain` and the HSV statistical value within the face bounding box to better detect skin color and prevent false detections.

[0255] Example 2

[0256] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a screen adjustment device disclosed in an embodiment of the present invention. Figure 3 As shown, the image adjustment device includes an acquisition module 310, a scene recognition module 320, an object recognition module 330, and an image adjustment module 340, wherein:

[0257] The acquisition module 310 is used to acquire the currently displayed screen;

[0258] The scene recognition module 320 is used to perform scene recognition processing on the current display screen to obtain the scene recognition result corresponding to the current display screen;

[0259] The object recognition module 330 is used to perform object recognition processing on the current display screen to obtain the object recognition result corresponding to the current display screen. The object recognition result is used to indicate whether there is a target object in the current display screen.

[0260] The image adjustment module 340 is used to adjust the image quality based on the object recognition result and the scene recognition result, and to adjust the color of the target object based on the object recognition result and the scene recognition result.

[0261] In one possible implementation, the scene recognition result includes the scene category corresponding to the currently displayed screen and the initial probability corresponding to each scene category, and the screen adjustment module 340 includes:

[0262] A probability smoothing unit is used to perform probability smoothing processing based on the object recognition result and the initial probability corresponding to each scene category to obtain the target probability corresponding to each scene category.

[0263] The parameter acquisition unit is used to acquire the image quality parameters corresponding to each of the scene categories;

[0264] The first adjustment unit is used to determine the target image quality parameters based on the image quality parameters corresponding to each scene category and the target probability corresponding to each scene category; and to adjust the image quality of the image based on the target image quality parameters.

[0265] In one possible implementation, the probabilistic smoothing unit includes:

[0266] The temporal filtering and smoothing subunit is used to perform temporal filtering on the initial probabilities corresponding to each of the scene categories to obtain the temporal filtered probabilities corresponding to each of the scene categories.

[0267] The priority suppression subunit is used to determine the priority corresponding to each scene category, and suppress the temporal filtering probability corresponding to the second scene category according to the priority corresponding to each scene category through the temporal filtering probability corresponding to the first scene category in the scene category, so as to obtain the candidate probability corresponding to each scene category, wherein the priority corresponding to the first scene category is higher than the priority corresponding to the second scene category.

[0268] The change clamping subunit is used to, if the object recognition result indicates that the target object exists in the current display screen, take the candidate probability corresponding to the target scene category associated with the target object as the target probability corresponding to the target scene category, and perform probability change clamping processing on the candidate probabilities corresponding to non-target scene categories to obtain the target probability corresponding to the non-target scene category. The non-target scene category includes categories other than the target scene category in the scene category.

[0269] In one possible implementation, the change clamping subunit is used to determine the probability change value of the non-target scene category based on the candidate probability corresponding to the non-target scene category and the initial probability corresponding to the non-target scene category;

[0270] Determine whether the probability change value is greater than the probability change threshold;

[0271] If the probability change value is less than or equal to the probability change threshold, then the candidate probability corresponding to the non-target scene category is taken as the target probability corresponding to the non-target scene category.

[0272] If the probability change value is greater than the probability change threshold, then the target probability corresponding to the non-target scene category is determined based on the initial probability corresponding to the non-target scene category and the probability change threshold. The target probability corresponding to the non-target scene category is positively correlated with the initial probability corresponding to the non-target scene category and the probability change threshold.

[0273] In one possible implementation, the priority suppression subunit is used to determine the maximum prediction probability, which is the maximum value among the temporal filtered probabilities corresponding to the first scene category;

[0274] The inhibition factor is determined by multiplying the maximum predicted probability by the preset inhibition coefficient.

[0275] The candidate probability corresponding to the second scene category is determined based on the difference between the temporal filtering probability and the suppression probability corresponding to the second scene category. The suppression probability is obtained by multiplying the temporal filtering probability corresponding to the second scene category with the suppression factor.

[0276] In one possible implementation, the scene recognition result includes the scene category corresponding to the current display screen and the initial probability corresponding to each scene category. The object recognition result is also used to represent the current position information of the target object in the current display screen. The screen adjustment module 340 further includes:

[0277] A probability acquisition unit is used to determine the target scene category associated with the target object, and to acquire the initial probability corresponding to the target scene category according to the initial probability corresponding to each scene category;

[0278] The weight determination unit is used to determine the target weight parameters for color adjustment of the target object based on the initial probability corresponding to the target scene category.

[0279] The HSV statistics unit is used to determine the target area occupied by the target object in the current display screen based on the current location information, and to count the HSV color information of the target area;

[0280] The second adjustment unit is used to adjust the color of the target object according to the target weight parameter and the HSV color information.

[0281] In one possible implementation, the weight determination unit is used to determine a scene reference weight based on the initial probability corresponding to the target scene category, wherein the scene reference weight is positively correlated with the initial probability corresponding to the target scene category;

[0282] A color reference weight is determined based on the object recognition result. The first color reference weight is greater than the second color reference weight. The first color reference weight is the weight corresponding to the presence of the target object in the current display screen, and the second color reference weight is the weight corresponding to the absence of the target object in the current display screen.

[0283] The target weight parameter is determined based on the scene reference weight and the color reference weight, and the target weight parameter is positively correlated with the scene reference weight and the color reference weight, respectively.

[0284] In one possible implementation, the current display screen is one video frame of the video stream, and the HSV statistics unit is used to obtain N reference display screens, which include screens that were displayed before the current display screen and where the target object exists, and N is a positive integer greater than or equal to 1.

[0285] For each frame of the reference display image, position smoothing processing is performed to obtain at least one candidate position information;

[0286] The target region is determined based on the at least one candidate location information;

[0287] The position smoothing process includes:

[0288] The IOU value of the target object is determined based on the current area occupied by the target object in the current display screen and the reference area occupied by the target object in the reference display screen.

[0289] If the IOU value is less than the IOU threshold, the candidate position information is determined based on the current position information and the average value of the reference position information of the target object in the reference display screen.

[0290] If the IOU value is greater than or equal to the IOU threshold, the candidate position information is determined based on the weighted average of the current position information and the reference position information of the target object in the reference display screen, wherein the reference display screen with more video frames between it and the current display screen has a smaller weight.

[0291] Example 3

[0292] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of another screen adjustment device disclosed in an embodiment of the present invention. For example... Figure 4 As shown, the screen adjustment device may include:

[0293] Memory 401 storing executable program code;

[0294] Processor 402 coupled to memory 401;

[0295] The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the screen adjustment method described in Embodiment 1 of the present invention.

[0296] Example 4

[0297] This invention discloses a computer-storable medium that stores computer instructions. When these computer instructions are invoked, they are used to execute the steps in the screen adjustment method described in Embodiment 1 of this invention.

[0298] Example 5

[0299] This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the screen adjustment method described in Embodiment 1.

[0300] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0301] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0302] Finally, it should be noted that the screen adjustment method and apparatus disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for adjusting screen display, characterized in that, The method includes: Get the currently displayed screen; The current display screen is subjected to scene recognition processing to obtain the scene recognition result corresponding to the current display screen; The current display screen is subjected to object recognition processing to obtain the object recognition result corresponding to the current display screen. The object recognition result is used to indicate whether the target object exists in the current display screen. The image quality is adjusted based on the object recognition result and the scene recognition result, and the color of the target object is adjusted based on the object recognition result and the scene recognition result; Furthermore, the scene recognition result includes the scene category corresponding to the currently displayed screen and the initial probability corresponding to each scene category. The step of adjusting the image quality based on the object recognition result and the scene recognition result includes: Based on the object recognition results and the initial probabilities corresponding to each scene category, probability smoothing is performed to obtain the target probability corresponding to each scene category. Obtain the image quality parameters corresponding to each of the aforementioned scene categories; The target image quality parameters are determined based on the image quality parameters corresponding to each scene category and the target probability corresponding to each scene category. Adjust the image quality according to the target image quality parameters; And, the step of performing probability smoothing processing based on the object recognition result and the initial probabilities corresponding to each scene category to obtain the target probability corresponding to each scene category includes: The initial probabilities corresponding to each scene category are subjected to temporal filtering to obtain the temporal filtered probabilities corresponding to each scene category. The priority of each scene category is determined, and the temporal filtering probability of the second scene category is suppressed by the temporal filtering probability of the first scene category in the scene category according to the priority of each scene category, so as to obtain the candidate probability of each scene category. The priority of the first scene category is higher than the priority of the second scene category. If the object recognition result indicates that the target object exists in the currently displayed screen, then the candidate probability corresponding to the target scene category associated with the target object is taken as the target probability corresponding to the target scene category, and the candidate probability corresponding to the non-target scene category is subjected to probability change clamping processing to obtain the target probability corresponding to the non-target scene category. The non-target scene category includes categories other than the target scene category in the scene category.

2. The method according to claim 1, characterized in that, The step of performing probability change clamping processing on the candidate probabilities corresponding to non-target scene categories to obtain the target probabilities corresponding to the non-target scene categories includes: Based on the candidate probability corresponding to the non-target scene category and the initial probability corresponding to the non-target scene category, determine the probability change value of the non-target scene category; Determine whether the probability change value is greater than the probability change threshold; If the probability change value is less than or equal to the probability change threshold, then the candidate probability corresponding to the non-target scene category is taken as the target probability corresponding to the non-target scene category. If the probability change value is greater than the probability change threshold, then the target probability corresponding to the non-target scene category is determined based on the initial probability corresponding to the non-target scene category and the probability change threshold. The target probability corresponding to the non-target scene category is positively correlated with the initial probability corresponding to the non-target scene category and the probability change threshold.

3. The method according to claim 1, characterized in that, The step of suppressing the temporal filtering probability corresponding to the second scene category in the scene category using the temporal filtering probability corresponding to the first scene category in the scene category includes: Determine the maximum prediction probability, which is the maximum value among the temporal filtered probabilities corresponding to the first scene category; The inhibition factor is determined by multiplying the maximum predicted probability by the preset inhibition coefficient. The candidate probability corresponding to the second scene category is determined based on the difference between the temporal filtering probability and the suppression probability corresponding to the second scene category. The suppression probability is obtained by multiplying the temporal filtering probability corresponding to the second scene category with the suppression factor.

4. The method according to any one of claims 1-3, characterized in that, The scene recognition result includes the scene category corresponding to the current display screen and the initial probability corresponding to each scene category. The object recognition result is also used to represent the current position information of the target object in the current display screen. The step of adjusting the color of the target object according to the object recognition result and the scene recognition result includes: Determine the target scene category associated with the target object, and obtain the initial probability corresponding to the target scene category based on the initial probability corresponding to each scene category; The target weight parameter for color adjustment of the target object is determined based on the initial probability corresponding to the target scene category. Based on the current location information, determine the target area occupied by the target object in the current display screen, and statistically analyze the HSV color information of the target area; The color of the target object is adjusted according to the target weight parameters and the HSV color information.

5. The method according to claim 4, characterized in that, The step of determining the target weight parameter for color adjustment of the target object based on the initial probability corresponding to the target scene category includes: A scene reference weight is determined based on the initial probability corresponding to the target scene category, and the scene reference weight is positively correlated with the initial probability corresponding to the target scene category. A color reference weight is determined based on the object recognition result. The first color reference weight is greater than the second color reference weight. The first color reference weight is the weight corresponding to the presence of the target object in the current display screen, and the second color reference weight is the weight corresponding to the absence of the target object in the current display screen. The target weight parameter is determined based on the scene reference weight and the color reference weight, and the target weight parameter is positively correlated with the scene reference weight and the color reference weight, respectively.

6. The method according to claim 4, characterized in that, The currently displayed frame is a video frame from a video stream. Determining the target area occupied by the target object in the currently displayed frame based on the current position information includes: Obtain N frames of reference display images, wherein the reference display images include images that were displayed before the current display image and contain the target object, and N is a positive integer greater than or equal to 1; For each frame of the reference display image, position smoothing processing is performed to obtain at least one candidate position information; The target region is determined based on the at least one candidate location information; The position smoothing process includes: The IOU value of the target object is determined based on the current area occupied by the target object in the current display screen and the reference area occupied by the target object in the reference display screen. If the IOU value is less than the IOU threshold, the candidate position information is determined based on the current position information and the average value of the reference position information of the target object in the reference display screen. If the IOU value is greater than or equal to the IOU threshold, the candidate position information is determined based on the weighted average of the current position information and the reference position information of the target object in the reference display screen, wherein the reference display screen with more video frames between it and the current display screen has a smaller weight.

7. A screen adjustment device, characterized in that, The device includes: The acquisition module is used to acquire the currently displayed screen. The scene recognition module is used to perform scene recognition processing on the current display screen to obtain the scene recognition result corresponding to the current display screen; An object recognition module is used to perform object recognition processing on the current display screen to obtain an object recognition result corresponding to the current display screen. The object recognition result is used to indicate whether a target object exists in the current display screen. The image adjustment module is used to adjust the image quality based on the object recognition result and the scene recognition result, and to adjust the color of the target object based on the object recognition result and the scene recognition result; Furthermore, the scene recognition result includes the scene category corresponding to the currently displayed screen and the initial probability corresponding to each scene category, and the screen adjustment module includes: A probability smoothing unit is used to perform probability smoothing processing based on the object recognition result and the initial probability corresponding to each scene category to obtain the target probability corresponding to each scene category. The parameter acquisition unit is used to acquire the image quality parameters corresponding to each of the scene categories; The first adjustment unit is used to determine the target image quality parameters based on the image quality parameters corresponding to each scene category and the target probability corresponding to each scene category; and to adjust the image quality of the image based on the target image quality parameters. And, a probability smoothing unit, including: The temporal filtering and smoothing subunit is used to perform temporal filtering on the initial probabilities corresponding to each of the scene categories to obtain the temporal filtered probabilities corresponding to each of the scene categories. The priority suppression subunit is used to determine the priority corresponding to each scene category, and suppress the temporal filtering probability corresponding to the second scene category according to the priority corresponding to each scene category through the temporal filtering probability corresponding to the first scene category in the scene category, so as to obtain the candidate probability corresponding to each scene category, wherein the priority corresponding to the first scene category is higher than the priority corresponding to the second scene category. The change clamping subunit is used to, if the object recognition result indicates that the target object exists in the current display screen, take the candidate probability corresponding to the target scene category associated with the target object as the target probability corresponding to the target scene category, and perform probability change clamping processing on the candidate probabilities corresponding to non-target scene categories to obtain the target probability corresponding to the non-target scene category. The non-target scene category includes categories other than the target scene category in the scene category.

8. A screen adjustment device, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the screen adjustment method as described in any one of claims 1-6.

Images