Portrait Mode Video Noise Reduction via Region-Specific Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video noise suppression technologies face challenges in ultra-low-brightness environments with strong noise intensity, where motion detection is difficult to distinguish between real object movement and noise, leading to suboptimal noise reduction, and high-resolution images struggle with real-time computing due to excessive computations.
Innovation Solution
A video noise reduction method and system that utilizes a portrait mode model to generate a current portrait probability image, performs binarization and contour extraction to differentiate foreground and background regions, and applies time-domain or spatial domain filtering based on motion detection, compensating motion regions with similar parts from previous images to reduce noise effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If time-domain noise suppression is performed on the current image according to previous image information, then noise reduction is achieved, but ghosting effect occurs for moving objects
Solution Approach 1:
The image is segmented into moving object regions and stationary background regions through motion detection. Different noise suppression strategies are applied to each segment: spatial domain filtering for moving regions (avoiding ghosting) and time-domain filtering for stationary regions (maximizing noise reduction). This segmentation resolves the contradiction by applying the appropriate filtering method to each region type.
Solution Approach 2:
Different filtering characteristics are applied to different regions of the image based on their motion properties. Moving regions receive spatial domain filtering with preserved temporal information, while stationary regions receive aggressive time-domain filtering. This local differentiation allows optimal noise suppression for each region type without causing ghosting artifacts.
2Object-generated harmful factors
If motion detection is performed to identify moving objects, then ghosting effect is avoided, but inability to distinguish real movement from noise occurs in ultra-low-brightness environments
Solution Approach 1:
The system dynamically adjusts the motion detection threshold based on image brightness conditions. In ultra-low-brightness environments with high noise levels, the threshold is raised to prevent false motion detection. This dynamic adaptation allows the system to maintain accurate motion detection capability while operating in challenging lighting conditions.
Solution Approach 2:
A confidence threshold mechanism is introduced as an intermediary between motion detection and filtering application. Only regions that exceed both the motion detection threshold and the confidence threshold undergo time-domain filtering. This intermediary layer prevents noise-induced false detections from triggering inappropriate filtering operations.
3Object-affected harmful factors
If high-resolution images are processed with neural network models, then noise suppression level is improved, but real-time computing becomes difficult due to excessive computations
Solution Approach 1:
The image processing is segmented into multiple stages: motion detection and region classification, selective filtering application, and confidence-based validation. This segmentation allows the system to process only relevant regions with appropriate algorithms, reducing overall computational load while maintaining noise suppression effectiveness.
Solution Approach 2:
Instead of applying full neural network processing to the entire high-resolution image, the system applies selective spatial and temporal filtering only to identified moving and stationary regions. This partial action approach achieves sufficient noise suppression for the specific application context while dramatically reducing computational requirements for real-time processing.
Data Source
AI summary
A video noise reduction method and a video noise reduction system based on a portrait mode are provided. The method includes configuring a processor to perform: inputting a current image into a portrait mode model to generate a current portrait probability image; executing a post-processing process, including performing a binarization process on the current portrait probability image to generate a mask image and executing a contour extraction process to extract a foreground region, a background region and a contour region; performing a filtering process, including: executing a motion detection process to locate a motion region and a stationary region; performing a time-domain filtering operation on the stationary region; for the motion region, obtaining a similar part of the previous image and a degree of difference of the similar part, so as to determine whether to compensate the motion region or to perform a spatial domain filtering operation.