Image Coding Boundary Filtering for Transform Unit Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image coding standards face inefficiencies in coding efficiency due to the difficulty in collecting difference signals into a low frequency range, especially when block sizes for frequency transform and motion compensation differ, leading to decreased data reduction and coding performance.
Innovation Solution
An image coding method that performs prediction processing on prediction units partitioned from target blocks, detects boundaries within transform units, applies boundary filtering, and calculates differences between prediction and input images to generate a difference image for frequency transform, allowing for smoother variations and extended transform unit sizes without internal boundaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If the block size for frequency transform is set larger than the block size for motion compensation, then the transform unit can cover a larger area for better correlation utilization, but a boundary of the motion compensation block is included in the transform block which generates steep edges that prevent effective low-frequency collection
Solution Approach 1:
The transform block is divided into multiple regions based on motion compensation block boundaries. Different processing methods are applied to regions containing boundaries versus regions without boundaries, allowing the system to maintain large transform unit sizes while handling boundary-induced steep edges separately through selective filtering or processing strategies.
Solution Approach 2:
Different quality processing is applied to different regions within the transform block. Regions containing motion compensation boundaries receive special handling (such as boundary detection and selective filtering) while regions without boundaries undergo standard transform processing, optimizing overall coding efficiency by addressing local issues without compromising global transform unit size.
2Ease of operation
If motion compensation and frequency transform are performed on small blocks (4x4 pixels), then processing can be done on fine-grained units, but correlation can only be used within limited space making it difficult to collect difference signals into low frequency range
Solution Approach 1:
Multiple small motion compensation blocks are merged into a larger transform block for frequency transformation. This combining approach allows motion compensation to operate on fine-grained 4x4 blocks while the subsequent transform operates on a larger unified block, enabling both fine processing control and effective low-frequency signal collection across a broader spatial range.
3Adaptability or versatility
If a macroblock is partitioned into small blocks for motion compensation, then motion compensation can be performed on each block with different motion vectors, but the transform block size becomes limited reducing correlation effectiveness
Solution Approach 1:
The macroblock is segmented into multiple motion compensation blocks for flexible motion estimation, while the transform operation spans across a larger combined area. This segmentation allows each sub-block to have its own motion vector for adaptability, while the unified transform block maintains larger size for better correlation and low-frequency collection.
Data Source
AI summary
An image coding method for coding an input image per block to generate a coded image signal includes: predicting for each prediction unit which is an area obtained by partitioning a target block to generate a prediction image of the target block; comparing a transform unit which is an area obtained by partitioning the target block and is a processing unit for frequency transform with the prediction unit, to detect part of a boundary of the prediction unit, the boundary being located within the transform unit; performing boundary filtering on the detected part of the boundary in the generated prediction image; calculating a difference between a filtered prediction image and the input image to generate a difference image of the target block; and performing frequency transform on the difference image for each transform unit.