Method for generating personalized poster based on intelligent viewing equipment

Abstract

The invention discloses a personalized poster generation method based on intelligent viewing equipment, and the method comprises the steps: taking a camera as a picture collection end, taking a display screen as a picture output end, and achieving the full-automatic generation of a personalized poster through the integration of intelligent matting, multi-style image processing and automatic synthesis technologies. The user only needs to upload the photo and shoot the scenery, the system can automatically complete portrait extraction, stylization processing and poster synthesis, high-quality and multi-style personalized posters are generated, and the user experience and sharing willingness are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing, and in particular to a personalized poster generation method based on an intelligent viewing device. BACKGROUND

[0002] With the development of the integration of tourism and digital multimedia technology, various types of scenic spots and viewing platforms have widely deployed intelligent viewing devices, such as digital telescopes and interactive photo booths. The core functions of these devices usually include collecting real-time scenery pictures through a camera and presenting them to tourists through a display screen.

[0003] However, the existing viewing devices have obvious technical bottlenecks and functional defects in providing personalized image services:

[0004] Lack of portrait fusion capability: Traditional devices mostly only provide simple scenery photography or simple picture superimposition. If a user wants to synthesize a personal image with a real scenery, they usually need to rely on complex post-processing software (such as Photoshop) for manual cutout and synthesis. This process is not only tedious and technically demanding, but also cannot be completed immediately on the viewing site, which seriously damages the continuity and experience of the tour.

[0005] Low system integration and automation: The complete process from image acquisition, portrait processing to final poster generation is usually scattered and non-integrated in the existing technology. Users need to switch between different applications or devices, and the operation process is broken. There is a lack of a terminal device integrated with a high-performance AI processing unit (such as NPU / GPU) to complete the one-stop full-process automatic processing from intelligent cutout, stylization to multi-layer automatic synthesis.

[0006] Therefore, there is an urgent need in the art for a technical solution that integrates an intelligent viewing device terminal, can automatically implement high-precision portrait cutout, and can synthesize a personalized poster with multiple style scenery photos, to overcome the shortcomings of the existing technology. SUMMARY

[0007] The present application proposes a personalized poster generation method based on an intelligent viewing device to solve the above-mentioned problems of the prior art.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical solution:

[0009] The present application provides a personalized poster generation method based on an intelligent viewing device, which uses a camera as a picture acquisition terminal and a display screen as a picture output terminal, and includes the following steps:

[0010] S1, the user selects and uploads an image to the intelligent viewing device through a mobile terminal, and sets its position and size on the display screen;

[0011] S2, the device end removes the background of the picture by an intelligent matting algorithm, only retaining the user portrait;

[0012] S3, after the user takes a landscape photo on the device end, the device first performs gray value processing on the taken landscape photo to generate photos of different styles;

[0013] S4, the device superimposes the user portrait and a blank poster on the basis of the landscape photo to generate a personalized poster;

[0014] The intelligent viewing device at least includes a camera for picture acquisition, a display screen capable of displaying the acquired picture to the tourists for browsing, and a core circuit board composed of a CPU, a GPU or an NPU processing chip to process the camera data.

[0015] Further, the specific process of step S1 is as follows:

[0016] After the device receives the user's picture, the coordinates X and Y of the lower left corner of the picture and the scaling ratio S are determined according to the position and size of the user-specified picture, and the scaled B picture is obtained.

[0017] Further, the specific process of step S2 is as follows:

[0018] According to the scaled B picture of step S1, the MODNet algorithm based on deep learning is used to perform portrait matting on the image, the pixel values of other areas except the user portrait are set to be transparent, and a portrait picture with only the user portrait and a transparent background is formed.

[0019] Further, the MODNet algorithm based on deep learning is used to perform portrait matting on the image in step S2, and three loss functions are used to optimize the model during the training process:

[0020] Semantic estimation loss: used to guide the model to roughly locate the portrait area, and provide support for focusing on the foreground area for subsequent processing;

[0021]

[0022] wherein S ij represents the value of the semantic map output by the network at (i, j); represents the label of the real semantic at (i, j); H s represents the height of the semantic map; W s represents the width of the semantic map;

[0023] Detail prediction loss:

[0024]

[0025] wherein Dij This represents the value of the detail plot at (i, j); H represents the true value of alpha at (i, j); d W represents the height of the detail image. d Represents the width of the detail image;

[0026] Fusion loss:

[0027]

[0028] Where F ij This represents the final predicted value at (i, j); H represents the true alpha value at (i, j); H represents the height of the image; W represents the width of the image.

[0029] The sum of the three loss functions mentioned above constitutes the overall loss function of deep learning. By iterating continuously to minimize the loss function, we can obtain the portrait image after removing the background.

[0030] Furthermore, the specific process of step S3 is as follows:

[0031] After the user takes a photo, a raw color image in JPG format is obtained. Each pixel in the raw color image has three pixel values: B, G, and R. First, sRGB gamma correction is performed:

[0032]

[0033] Where C sRGB Divide the original pixel value by 255, C linear These are linear pixels after gamma removal;

[0034] Calculate linear luminance:

[0035] Y = 0.2126 × R linear +0.7152×G linear 0.0722×B linear

[0036] Where Y is linear luminance, R linear G linear B linear These correspond to the linear pixel values ​​of the red, green, and blue components of the pixel calculated in the previous step;

[0037] Re-gamma compression:

[0038]

[0039] Where L linear L is the linear brightness obtained in the previous step. sRGB Divide by 255 and round down to get the final grayscale value.

[0040] Furthermore, the specific process of step S4 is as follows:

[0041] First, the image processing tool ffmpeg overlays the transparent background user portrait obtained in step S2 onto the color and grayscale landscape images generated in step S3, and then further overlays them onto the background poster, ultimately forming two different styles of personalized posters.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] This invention proposes a method for generating personalized posters based on intelligent viewing devices. By integrating intelligent image cutout, multi-style image processing, and automatic compositing technologies, it achieves fully automated personalized poster generation. Users only need to upload photos and capture scenery; the system can automatically extract portraits, stylize them, and composite the posters, generating high-quality, multi-style personalized posters, effectively enhancing user experience and encouraging sharing. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 A flowchart illustrating a method for generating personalized posters based on intelligent viewing devices, as provided in an embodiment of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of 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.

[0047] This invention provides a method for generating personalized posters based on intelligent viewing devices, using a camera as the image acquisition end and a display screen as the image output end, such as... Figure 1 As shown, it includes the following steps:

[0048] S1. Users select and upload images to the smart viewing device via mobile terminal (such as mobile phone), and set its position and size on the display screen.

[0049] The intelligent viewing device includes at least a camera for image capture, a display screen for showing the captured images to tourists, and a core circuit board composed of CPU, GPU or NPU processing chips to process camera data.

[0050] The specific process of step S1 is as follows:

[0051] When a user selects a photo on a mobile device (such as a mobile phone), the mobile device will first scale the photo proportionally and place it in a 1920*1080 frame. The user can adjust the position and size of the photo in the frame. After clicking "start projection", the image at that position and size will be automatically synchronized to the screen.

[0052] After receiving the user's image, the device determines the X and Y coordinates of the lower left corner of the image, as well as the scaling ratio S, based on the user-specified image position and size, to obtain the scaled B image.

[0053] S2. The device uses an intelligent background removal algorithm to remove the background of the image, leaving only the user profile.

[0054] The specific process of step S2 is as follows:

[0055] Based on the scaled B image from step S1, the MODNet algorithm based on deep learning is used to perform portrait matting on the image, setting the pixel values ​​of areas other than the user portrait to transparent, thus forming a portrait image where only the user portrait is transparent and the background is transparent.

[0056] Furthermore, step S2 utilizes the MODNet algorithm based on deep learning to perform portrait matting on the image. During training, three loss functions are used to jointly optimize the model:

[0057] Semantic estimation loss: used to guide the model to roughly locate the human image region, providing support for subsequent processing to focus on the foreground region;

[0058]

[0059] Where S ij This represents the value of the semantic graph output by the network at (i, j); The label representing the actual semantic meaning at (i, j); H s W represents the height of the semantic graph. s Represents the width of the semantic graph;

[0060] Detail prediction loss: Accurately restore alpha matte under high resolution conditions, focusing on improving the restoration quality of portrait edge details;

[0061]

[0062] Where D ij This represents the value of the detail plot at (i, j); H represents the true value of alpha at (i, j); d W represents the height of the detail image. d Represents the width of the detail image;

[0063] Fusion loss: It helps the predicted alpha image to maintain consistency with the real image in terms of edge structure, and is particularly helpful in preserving high-frequency details such as hair strands and semi-transparent areas;

[0064]

[0065] Where F ij This represents the final predicted value at (i, j); H represents the true alpha value at (i, j); H represents the height of the image; W represents the width of the image.

[0066] The sum of the three loss functions mentioned above constitutes the overall loss function of deep learning. By iterating continuously to minimize the loss function, a portrait image with clean background removal and complete detail preservation can be obtained.

[0067] S3. After the user takes a landscape photo on the device, the device first processes the landscape photo with grayscale to generate photos of different styles.

[0068] The specific process of step S3 is as follows:

[0069] After a user takes a photo, they receive an original color image in JPG format. To enhance the historical feel and artistic effect of the photo, the system performs grayscale processing on the JPG image. The specific steps are as follows:

[0070] Each pixel in the original color image has three pixel values: B, G, and R. First, sRGB gamma correction is performed:

[0071]

[0072] Where C sRGB Divide the original pixel value by 255, C linear These are linear pixels after gamma removal;

[0073] Calculate linear luminance:

[0074] Y = 0.2126 × R linear +0.7152×G linear 0.0722×B linear

[0075] Where Y is linear luminance, R linear Glinear B linear These correspond to the linear pixel values ​​of the red, green, and blue components of the pixel calculated in the previous step;

[0076] Re-gamma compression:

[0077]

[0078] Where L linear L is the linear brightness obtained in the previous step. sRGB Divide by 255 and round down to get the final grayscale value.

[0079] S4. The device overlays user profiles and blank posters onto landscape photos to generate personalized posters.

[0080] The specific process of step S4 is as follows:

[0081] First, using image processing tools such as ffmpeg, the transparent background user portrait obtained in step S2 is overlaid onto the two landscape images (one in color and one in grayscale) generated in step S3. Then, it is further overlaid onto the background poster, ultimately creating two personalized posters in different styles (color and grayscale). Users can download the generated posters to their phones and share them on social media platforms (such as WeChat Moments).

[0082] The present invention provides a method for generating personalized posters based on intelligent viewing devices. Without any additional human intervention, after the user uploads an image, the device automatically cuts out the image and overlays it with the user's photo and background poster to generate a personalized poster. After downloading it to a mobile terminal, tourists can share it on their social media, satisfying their pursuit of personalization and sharing needs.

[0083] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1.A method for generating a personalized poster based on an intelligent viewing device, characterized in that, A camera is used as a picture acquisition terminal, and a display screen is used as a picture output terminal, and the method comprises the following steps: S1, a user selects and uploads an image to an intelligent viewing device through a mobile terminal, and sets the position and size of the image on the display screen; S2, the device removes the background of the image through an intelligent matting algorithm, and only retains the user's portrait; S3, after the user takes a picture of a scenic spot on the device, the device first performs gray value processing on the taken picture of the scenic spot to generate pictures of different styles; S4, the device superimposes the user's portrait and a blank poster on the basis of the picture of the scenic spot to generate a personalized poster. The intelligent viewing device comprises at least a camera for picture acquisition, a display screen capable of displaying the acquired picture to a tourist for browsing, and a core circuit board composed of a CPU, a GPU or an NPU for processing camera data. 2.The method of claim 1, wherein, The specific process of step S1 is as follows: After receiving the user's picture, the device determines the coordinates X and Y of the lower left corner of the picture and a scaling ratio S according to the position and size of the picture specified by the user, and obtains a scaled B picture. 3.The method of claim 1, wherein, The specific process of step S2 is as follows: According to the scaled B picture of step S1, the MODNet algorithm based on deep learning is used to perform portrait matting on the image, and the pixel values of other areas except the user's portrait are set to be transparent to form a portrait picture with only the user's portrait and a transparent background. 4.The method of claim 3, wherein, In the training process, three loss functions are used to optimize the model together: Semantic estimation loss: where S ij the value of the semantic graph representing the network output at (i, j); the label of the true semantics at (i, j); H s the height of the semantic graph; W s the width of the semantic graph; Detail prediction loss: where D ij represents the value of the detail map at (i, j); represents the value of the true alpha at (i, j); H d represents the height of the detail map; W d represents the width of the detail map; Fusion loss: where F ij represents the final prediction value at (i, j); represents the value of the true alpha at (i, j); H represents the height of the image; W represents the width of the image; The above three loss functions are superimposed to form the entire deep learning loss function, and the loss function is minimized through continuous iteration, so that the portrait picture after background removal is obtained. 5.The method of claim 1, wherein, The specific process of step S3 is as follows: After the user takes a picture, a raw color picture in jpg format is obtained, and each pixel point of the raw color picture has three pixel values of B, G and R. First, sRGB gamma correction is performed: where C sRGB is the original pixel value divided by 255, C linear is the linear pixel after de-gamma. Linear luminance is calculated: Y = 0.2126 x R linear + 0.7152 x G linear 0.0722 x B linear where Y is linear luminance, R linear , G linear , and B linear correspond to the linear pixel values of the red, green, and blue components of the pixel calculated in the previous step, respectively. Re-gamma compression is performed: where L linear is the linear luminance of the previous step, resulting in L sRGB divided by 255 and rounded to the nearest integer is the final gray value. 6.The method of claim 1, wherein, The specific process of step S4 is as follows: First, the image processing tool ffmpeg is used to superimpose the transparent background user portrait obtained in step S2 on the color and grayscale scenic pictures generated in step S3, and then further superimpose it on the background poster to finally form two different styles of personalized posters.

Images