Privacy Mask Alignment with Encoding Units
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding methods often distort privacy masks during the encoding of video sequences, leading to unintended exposure of masked regions, particularly when faces or objects move across defined privacy zones.
Innovation Solution
A method that redefines privacy masks to align with encoding units and restricts prediction encoding within these masks, allowing only inter encoding between privacy masked regions of different frames, while enabling temporal prediction based on classification and privacy levels to maintain privacy and enhance encoding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If standard video encoding with prediction encoding is used, then encoding efficiency is improved, but privacy mask integrity deteriorates due to distortion and unintended exposure of masked regions
Solution Approach 1:
The image is divided into encoding units (e.g., 16x16 pixel blocks) and further segmented into masked and unmasked regions. The privacy mask is extended to align with encoding unit boundaries, ensuring that each encoding unit is fully masked or fully unmasked. This segmentation prevents prediction encoding from mixing masked and unmasked regions, thereby maintaining mask integrity while allowing efficient encoding within each segment.
Solution Approach 2:
Different encoding strategies are applied to different regions of the image based on their privacy requirements. Masked regions use temporal prediction encoding only (from previous masked regions), while unmasked regions use full spatial and temporal prediction encoding. This local differentiation maintains privacy protection in masked areas while maximizing encoding efficiency in unmasked areas.
2Productivity
If prediction encoding from outside privacy mask regions is allowed, then coding gain is improved, but privacy protection deteriorates due to contamination of masked regions
Solution Approach 1:
The harmful influence of unmasked regions on masked regions is extracted and removed by disabling spatial prediction encoding from unmasked to masked regions. The encoder is instructed to use only temporal prediction from previously encoded masked regions, completely separating the prediction sources to prevent any contamination of masked regions with unmasked region data.
Solution Approach 2:
Previously encoded masked regions serve as an intermediary for predicting current masked regions. Instead of allowing direct spatial prediction from unmasked regions, the system uses temporal prediction from previously masked regions as an intermediary step, ensuring that only safe, already-protected data influences the current masked region encoding.
3Manufacturing precision
If privacy mask is applied at pixel level, then privacy precision is improved, but encoding complexity increases due to misalignment with encoding units
Solution Approach 1:
The privacy mask boundaries are merged with encoding unit boundaries by extending the mask to cover complete encoding units. This alignment merges the two previously separate concepts (pixel-level mask precision and block-level encoding units) into a unified structure, eliminating the complexity of partial encoding unit masking while maintaining effective privacy protection.
Data Source
AI summary
A method, system, and non-transitory computer readable medium for encoding a privacy masked image into an encoded image frame of a video stream is disclosed. The encoding comprises receiving pixel divided image data of an image; receiving data defining an area and graphical characteristic of a privacy mask that masks part of the image; extending the privacy mask to be aligned with one or more encoding units having one or more pixels located within the privacy mask to form a redefined privacy mask; generating a privacy masked image by applying the redefined privacy mask and the graphical characteristic of the privacy mask to the image; and encoding the privacy masked image into the encoded image frame of the video stream, wherein the encoding comprises preventing spatial prediction encoding and/or temporal prediction encoding of pixels within the redefined privacy mask of the image in response to certain conditions.