Fast Integer Transform for Arbitrary Length Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression technologies face inefficiencies in handling arbitrary transform lengths and directional features, particularly with the Discrete Cosine Transform (DCT), which limits their ability to effectively compress images with edges aligned arbitrarily, leading to suboptimal energy compaction and compression performance.
Innovation Solution
The development of novel fast integer and directional transforms, including one-dimensional and two-dimensional transforms, which approximate the DCT without requiring factorization, allowing for arbitrary transform lengths and accommodating directional image features, thereby improving compression efficiency and energy compaction across various image orientations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional DCT is used for video compression, then energy compaction performance is optimal for images with large correlation coefficients, but the transform cannot effectively handle arbitrary transform lengths and directional features
Solution Approach 1:
The patent changes the fundamental parameters of the transform by developing fast integer transforms for arbitrary lengths (not just power of 2) and creating directional transforms that can adapt to different image orientations. This allows the transform to be flexible in length while maintaining good energy compaction by matching the transform direction to edge orientations in the image.
Solution Approach 2:
The patent introduces dynamic adaptability by allowing the transform length and direction to be selected based on the characteristics of the image block being processed. The directional transforms can be rotated to match edge orientations, making the system adaptive rather than static like traditional DCT.
2Adaptability or versatility
If fast integer transforms are designed for arbitrary lengths, then adaptability to different transform sizes is improved, but computational complexity increases
Solution Approach 1:
The patent segments the arbitrary length transform into smaller manageable components. The fast integer transform algorithms divide the transform computation into stages that can handle power-of-2 lengths efficiently, with additional handling for the remaining arbitrary length portions, thereby reducing overall computational complexity.
Solution Approach 2:
The patent introduces intermediary structures and algorithms that bridge between the efficient power-of-2 transform computations and the arbitrary length requirements. These intermediary fast integer transform algorithms act as mediators that maintain efficiency while achieving arbitrary length support.
3Productivity
If directional transforms are implemented to handle arbitrary orientations, then compression efficiency for images with arbitrary edge orientations is improved, but transform complexity increases
Solution Approach 1:
The patent applies asymmetry by creating directional transforms that are specifically optimized for different orientation directions. Instead of using a single symmetric DCT for all orientations, the system employs asymmetric directional transforms that are tailored to match specific edge orientations, improving compression efficiency for images with arbitrary edge directions.
4Speed
If rounding operations are used in fast integer transforms, then computational speed is improved, but precision is reduced
Solution Approach 1:
The patent incorporates feedback mechanisms in the form of correction terms and compensation operations that account for the precision loss introduced by rounding. The transform algorithms include feedback steps that adjust the rounded values to maintain higher precision while still benefiting from the speed of integer operations.
Data Source
AI summary
Methods, apparatuses and systems directed to frequency domain transforms, including fast integer transforms and directional integer transforms. Further described is a video codec that utilizes a novel class of fast integer and directional transforms.