Video Advertisement Embedding via Style Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in automatically detecting suitable positions for embedding advertisements in videos and seamlessly integrating them into the surrounding environment, leading to obtrusive and inefficient advertising.
Innovation Solution
A method utilizing a computer device equipped with convolutional neural networks and a generative adversarial network to automatically detect and position advertisements within videos, ensuring seamless integration by matching the advertisement's style with the surrounding video pixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional advertisement placement modes (pre-movie ads or ads during pauses) are used, then the implementation is simple, but the user experience deteriorates (obtrusive and blunt) and advertising conversion rate is low
Solution Approach 1:
The patent replaces conventional mechanical advertisement placement methods with deep learning-based automated detection and stitching systems. Convolutional neural networks automatically identify suitable advertisement positions and generate stitching masks, substituting manual or rule-based approaches with intelligent algorithms that understand video content semantics.
Solution Approach 2:
The patent transforms advertisement placement from fixed positional parameters (beginning or pause points) to dynamic content-based parameters. The system evaluates multiple dimensions including spatial location, temporal duration, content relevance, and visual compatibility to determine optimal placement, fundamentally changing the parameter space from simple time-position to multi-dimensional content awareness.
2Extent of automation
If automated advertisement detection methods based on simple feature parameters are used, then the automation level increases, but the detection precision and seamless integration capability deteriorate
Solution Approach 1:
The patent replaces simple feature-parameter-based detection with deep learning convolutional neural networks. The CNN architecture automatically learns complex spatial and temporal patterns from video content, substituting manual feature engineering with end-to-end learning that achieves superior detection precision without sacrificing automation.
Solution Approach 2:
The patent employs a composite technical approach combining multiple deep learning models (detection model, stitching mask generation model, style transfer model) working together. This composite system integrates different functional components that collectively achieve both high automation and high precision, with each model contributing specialized capabilities.
3Speed
If template matching methods are used for advertisement stitching, then the processing speed is faster, but the seamless integration quality deteriorates
Solution Approach 1:
The patent replaces template matching algorithms with generative adversarial networks (GANs) and style transfer models. These deep learning approaches generate photorealistic advertisement integrations that adapt to surrounding video content, substituting rigid template-based methods with flexible content-aware generation that maintains both speed and quality.
Solution Approach 2:
The patent transforms the stitching process from parameter-based template alignment to learned feature-space transformation. The style transfer model operates in a learned feature space rather than pixel space, enabling efficient processing through pre-trained network optimizations while achieving seamless visual integration that adapts to varying content characteristics.
4Manufacturing precision
If manual advertisement position identification is used, then the integration quality can be controlled, but the productivity and automation level deteriorate
Solution Approach 1:
The patent implements self-service through fully automated deep learning pipelines that perform detection, positioning, stitching mask generation, and style transfer without human intervention. The system serves itself by learning from data and automatically optimizing placement decisions, eliminating the need for manual position identification while maintaining or improving integration quality through learned patterns.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Embodiments of the present invention disclose a method for embedding an advertisement in a video and a computer device. The method includes: determining, by a computer device, a target image, where the target image is an image that is in M frames of images of a target video and that includes a first print advertisement, and M is a positive integer; determining, by the computer device, a target area, where the target area is an area in which the first print advertisement is located in the target image; inserting, by the computer device, a to-be-embedded second print advertisement into the target area to replace the first print advertisement; and converting, by the computer device, a style of the target image in which the second print advertisement is embedded, where a style of the second print advertisement in the target image after style conversion is consistent with a style of an image pixel outside the area in which the second print advertisement is located in the target image. By implementing the embodiments of the present invention, the computer device may fully automatically pinpoint an advertisement position, embed an advertisement, and keep consistent style for the embedded advertisement, to improve visual experience of the embedded advertisement.