Pixel Adaptive Intra Smoothing for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image block prediction methods in video coding, such as H.264/MPEG-4 AVC, face inefficiencies due to high frequency coefficients in prediction errors, particularly for non-linear edges or edges not matching predefined prediction directions, leading to increased bitrate or distortion.
Innovation Solution
Applying filtering to reference pixels based on their distance from the pixel to be predicted, using either smoothing or unfiltered reference pixels depending on their proximity, to reduce high frequency components in prediction errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If spatial prediction is applied using predefined prediction directions, then prediction speed and simplicity are improved, but prediction accuracy deteriorates for non-linear edges
Solution Approach 1:
The patent applies dynamic filtering where the filter strength is adjusted based on the distance between reference pixels and pixels to be predicted. Closer pixels receive stronger filtering to reduce high-frequency errors, while farther pixels receive weaker filtering to preserve edge details. This dynamic adaptation resolves the contradiction by making the prediction process both fast (through automated distance-based rules) and accurate (through adaptive filtering intensity).
Solution Approach 2:
The patent applies different filtering strengths to different spatial locations within the prediction block. Pixels near reference pixels receive stronger filtering while pixels farther away receive weaker filtering. This local differentiation allows the system to maintain high prediction accuracy in regions where it matters most (near edges) while preserving overall prediction speed through automated local adaptation.
2Object-generated harmful factors
If strong filtering is applied to reference pixels, then high frequency coefficients in prediction errors are reduced, but prediction accuracy for pixels near reference pixels deteriorates
Solution Approach 1:
The patent applies different filtering strengths to different spatial locations. Pixels close to reference pixels receive weaker filtering to preserve accuracy, while pixels farther from reference pixels receive stronger filtering to reduce high-frequency errors. This local differentiation resolves the contradiction by applying strong filtering only where it benefits accuracy (farther pixels) rather than uniformly across all pixels.
Solution Approach 2:
The filtering strength is dynamically adjusted based on the distance metric between reference pixels and pixels to be predicted. This dynamic adaptation allows the system to automatically reduce high-frequency coefficients in regions where they cause harm (farther pixels) while preserving accuracy in regions where reference pixels are most influential (closer pixels).
3Device complexity
If uniform filtering is applied to all pixels in the image block, then processing simplicity is improved, but prediction accuracy for pixels at different distances from reference pixels deteriorates
Solution Approach 1:
The patent implements dynamic, distance-based filtering that automatically adapts to each pixel's position relative to reference pixels. While the filtering strength varies per pixel (improving accuracy), the overall process remains relatively simple through automated distance calculations and threshold-based filtering decisions, avoiding the need for complex manual configuration.
Solution Approach 2:
The patent performs preliminary filtering of reference pixels before using them for prediction, based on pre-calculated distance metrics. This preliminary action prepares the reference data in advance, allowing the main prediction process to proceed efficiently while already having optimized the filtering strength according to spatial relationships, thus balancing simplicity and accuracy.
Data Source
AI summary
The present invention relates to spatial prediction of pixels of a block, the block being a block of a digital image. In particular, for a block pixel at least one reference pixel(s) is selected and out of the selected reference pixel(s), the block pixel is predicted. In particular, the prediction of the block pixel is performed depending on the distance of this pixel to the reference pixel(s) from which it is to be predicted.