SAO Edge Offset Processing With Unified Pixel Categorization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing edge offset (EO) mode in sample adaptive offset (SAO) technology requires reading a reconstructed pixel four times, leading to significant time consumption and performance degradation.
Innovation Solution
A method for data processing in the edge offset mode that involves dividing a target reconstructed image into non-overlapping coding tree blocks, traversing pixels to calculate residual values and categories, and using a single function to determine categories for all pixels in the edge offset modes, reducing the need for multiple functions and minimizing redundant calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If four separate functions are written for four EO directions to complete pixel category calculation, then the completeness of edge offset processing is improved, but the device complexity and processing time increase
Solution Approach 1:
The patent merges four separate edge offset processing functions (one for each direction: horizontal, vertical, 135°, and 45°) into a single unified function. This single function processes all four directions by traversing pixels in different orientations and calculating categories accordingly, thereby reducing code complexity while maintaining complete edge offset processing capability.
Solution Approach 2:
The unified edge offset processing function is designed with multi-functionality to handle all four edge offset directions within a single function implementation. The function universally processes pixels regardless of their orientation, making it a multi-functional replacement for four separate direction-specific functions.
2Reliability
If four separate functions are written for four EO directions, then the completeness of edge offset processing is improved, but the processing time increases
Solution Approach 1:
By combining four separate processing functions into one unified function, the patent reduces the total number of function calls and sequential processing steps. The single function processes all four directions in a more efficient manner, reducing overall processing time while maintaining complete edge offset processing.
Solution Approach 2:
The unified function performs preliminary actions by pre-determining the processing sequence and logic for all four directions before actual pixel processing begins. This allows for more efficient execution by avoiding redundant setup operations that would occur with four separate functions.
3Measurement precision
If multiple functions are used for different EO directions, then the accuracy of pixel category calculation is improved, but the computing time increases
Solution Approach 1:
The patent merges the accuracy-finding logic for all four edge offset directions into a single unified function. This single function maintains the same level of accuracy in pixel category calculation by using consistent comparison logic for all directions, while reducing computing time by eliminating redundant function calls and streamlining the processing flow.
Data Source
AI summary
Techniques for data processing in an edge offset mode of sample adaptive offset are provided. The techniques include: traversing reconstructed pixels in a coding tree block, and calculating a residual value between each reconstructed pixel and original pixel; traversing all first reconstructed pixels in a non-boundary part of the coding tree block, and calculating a category of each traversed first reconstructed pixel in each direction; traversing all second reconstructed pixels in a boundary part of the coding tree block, and calculating a category of each traversed second reconstructed pixel along a single direction; and calculating an accumulated residual value of reconstructed pixels along the four directions and an accumulated number of reconstructed pixels under the category based on the category of each first reconstructed pixel the edge offset modes along the four directions, the category of each second reconstructed pixel along the single direction, and each residual value.