Directional Discrete Cosine Transform for Image Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Discrete Cosine Transform (DCT) methods apply indiscriminately in vertical and horizontal directions, failing to effectively de-correlate image regions with strong directional properties, such as visual edges and textures, which are common in natural and computer-generated images.
Innovation Solution
The system performs directional DCT by determining the directional property of an image and factorizing DCT operations into primal operations, applying them along the determined direction, and using motion compensation for video sequences, optimizing coding modes between adjacent blocks through a weighted graph.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional DCT is applied indiscriminately in vertical and horizontal directions, then the transform can be applied uniformly to all image blocks, but it fails to effectively de-correlate image regions with strong directional properties
Solution Approach 1:
The patent applies dynamics by making the DCT direction adaptive rather than fixed. The transform direction changes dynamically based on the detected directional properties of different image regions. This allows the system to switch between horizontal, vertical, and diagonal DCT applications according to the local image characteristics, thereby improving de-correlation performance while maintaining ease of application through automated direction selection
Solution Approach 2:
The patent implements local quality by applying different DCT directions to different regions of the image based on their specific directional properties. Instead of using a uniform transform direction for the entire image, the system analyzes local directional characteristics and applies the most appropriate DCT orientation (horizontal, vertical, or diagonal) to each region, optimizing the transform performance for each local area
2Manufacturing precision
If one-dimensional DCT is applied along an arbitrary direction to improve performance for directional regions, then de-correlation performance improves, but it becomes not straightforward to perform the transform
Solution Approach 1:
The patent applies segmentation by dividing the image into multiple regions based on their directional properties, and applying the appropriate one-dimensional DCT to each segment. This segmentation approach allows the complex arbitrary direction transforms to be broken down into simpler, manageable one-dimensional transforms applied to specific image segments, reducing overall computational complexity while maintaining improved de-correlation performance
Solution Approach 2:
The patent utilizes parameter changes by rotating the image or transform coordinates to align with the dominant directional property of each region. By changing the orientation parameter of the DCT transform to match the local image structure, the system can apply standard one-dimensional DCT algorithms along arbitrary directions, improving performance without requiring entirely new transform methodologies
Data Source
AI summary
Systems and methods provide directional discrete cosine transformation (DCT) and motion compensated DCT. In one implementation, an exemplary system finds a directional property of an image, such as a visual trend, factorizes a DCT operation into primal operations, and applies the primal operations along a corresponding direction to perform the DCT. Motion compensated DCT applies the primal operations along a motion trajectory of a video sequence. When the directional DCT is applied blockwise, the directional coding modes for adjacent blocks can be optimized in view of each other using a weighted graph to represent the related coding mode decisions.