Video Information Embedding with Foreground-Preserving Masks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for embedding information into video backgrounds lack effectiveness in ensuring that foreground content is not occluded, leading to suboptimal viewing experiences.
Innovation Solution
A method is developed to construct a model based on pixel distribution characteristics of an embedding region in a reference frame, identify background and foreground, and generate a template to occlude the background while exposing the foreground, using techniques like Gaussian mixture modeling and feature point matching for static and motion shots.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If information is superposed in the background of the video, then information embedding is achieved, but foreground content may be occluded
Solution Approach 1:
The patent applies preliminary action by constructing a background model before embedding information. The system builds a statistical model of the background content in advance, then uses this model to identify and preserve foreground regions during the information embedding process. This allows the system to pre-determine which areas should not be occluded, preventing foreground loss before it occurs.
Solution Approach 2:
The patent uses a background model as an intermediary between the information embedding process and the foreground content. This statistical model acts as a mediator that guides the embedding process to avoid foreground regions. The model compares current frame pixels with the learned background model to identify foreground areas, thereby protecting them from occlusion while still allowing information embedding in safe background regions.
2Measurement precision
If a model is constructed based on pixel distribution characteristics, then foreground identification accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies parameter changes by using statistical parameters (mean and variance) to represent background pixel characteristics. Instead of storing and comparing every pixel value, the system models background pixels using Gaussian distributions defined by these statistical parameters. This reduces the computational burden while maintaining accurate foreground identification, as comparing against a statistical model is more efficient than against raw pixel data.
Solution Approach 2:
The patent applies local quality by constructing separate background models for different spatial regions of the video frame. Rather than using a single global background model, the system allows different regions to have different background characteristics. This improves foreground identification accuracy in heterogeneous scenes while managing computational complexity by only modeling regions that require it.
Data Source
Figure 1A~1B
Figure 2~3
Figure 4
AI summary
The present application provides a method for implanting information into a video, a computer device and a storage medium. The method comprises: constructing a model conforming to pixel distribution characteristics of an implantation region in a reference frame, and controlling the update of the model on the basis of a frame to be detected subsequent to the reference frame; recognizing, on the basis of the model, the background and the foreground of an implantation region in the frame to be detected, and generating a template for blocking the background and exposing the foreground; applying the template to information to be implanted, to block content of said information which may block the foreground; using said information, to which the template has been applied, to overlay the implantation region in the frame to be detected, so that the foreground is highlighted relative to said information. The present application efficiently integrates information into background content of a video.