Dynamic Background Image Generation for Privacy Protection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques for privacy protection face issues with background image updates, leading to hole-like defective states and excessive foreground extraction due to luminance changes, and fail to accurately separate moving objects from backgrounds.
Innovation Solution
An image processing system that includes an image acquisition unit, detection unit, background generation unit, and silhouette image generation unit, which detects person regions, generates a background image by combining non-foreground regions, and updates the background image using a weighted average of pixel values to minimize hole-like defects and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a background image is acquired at timing when no foreground is shown and used in a fixed manner, then the simplicity of the processing is improved, but the accuracy of foreground extraction deteriorates due to luminance changes causing excessive region extraction
Solution Approach 1:
The patent implements dynamic background image updating by combining multiple captured images over time. Instead of using a single fixed background image, the system continuously acquires multiple images and updates the background image to adapt to luminance changes in the environment, thereby maintaining accurate foreground extraction despite lighting variations.
Solution Approach 2:
The patent changes the parameter of background image temporal composition by acquiring and combining multiple images at different time points. This parameter change allows the system to distinguish between temporary luminance fluctuations and actual foreground objects, improving extraction accuracy while managing processing complexity through efficient image combination algorithms.
2Adaptability or versatility
If regions other than human body or moving object regions are combined to successively update a background image, then the adaptability to environment changes is improved, but the reliability of the background image deteriorates due to hole-like defective states
Solution Approach 1:
The patent applies preliminary action by detecting human bodies and moving objects before updating the background image. By identifying and excluding these regions in advance, the system prevents them from being incorrectly incorporated into the background image, thereby avoiding hole-like defective states and maintaining background image reliability while still adapting to environmental changes.
Solution Approach 2:
The patent implements feedback mechanisms by continuously monitoring detected human bodies and moving objects to adjust background image updates. The system uses detection results to provide feedback on which regions should or should not be updated, ensuring that the background image adapts to environmental changes without developing defects from persistent moving object regions.
3Measurement precision
If a background image is updated by combining multiple captured images, then the accuracy of foreground extraction is improved, but the processing time increases
Solution Approach 1:
The patent applies partial action by selectively combining only the necessary portions of multiple captured images for background updates, rather than processing entire images unnecessarily. This approach maintains high foreground extraction accuracy by focusing computational resources on relevant regions while reducing overall processing time through optimized image combination operations.
Data Source
Figure 1A~1B
Figure 2
Figure 3A~3D
AI summary
An image processing apparatus includes detection means configured to detect a first region in which a foreground object is present with respect to a plurality of captured images, retention means configured to retain a first background image, generation means configured to generate a second background image based on portions of each of the plurality of captured images which are not detected as a first region, and output means configured to select one of the first background image and the second background image based on a property of the second background image,, and configured to output, based on the selected background image and the first region, an image in which the foreground object is obscured.