Image Decoder Prediction Signal Generation for Overlapping Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image coding methods face inefficiencies in prediction when a prediction block and a coding block overlap, leading to reduced compression rates, especially in cases with periodic patterns or discontinuous images, where the prediction methods do not effectively utilize the correlation between pixels.
Innovation Solution
An image processing device and method that uses a prediction vector to generate a prediction signal by extracting pixel information from an already-decoded area at an integer multiple of the vector's distance, rather than relying on the overlapping area, thereby enhancing prediction efficiency and reducing the amount of codes required.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional prediction methods are used with overlapping prediction blocks, then the prediction process can be simplified, but the prediction efficiency decreases and code amount increases
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing pixel information from already-decoded areas at integer multiple distances before the actual prediction is needed. This allows the prediction process to efficiently retrieve pre-prepared data rather than performing complex real-time calculations, thereby improving prediction efficiency without significantly increasing device complexity
Solution Approach 2:
The patent segments the prediction process by dividing it into distinct steps: identifying overlapping areas, selecting pixel information at integer multiple distances from already-decoded regions, and generating prediction signals separately for overlapping and non-overlapping areas. This segmentation allows each part to be optimized independently, improving overall prediction efficiency
2Ease of operation
If pixel information from overlapping areas is used for prediction, then the prediction process is straightforward, but the amount of codes increases and compression rate decreases
Solution Approach 1:
The patent applies local quality by using different prediction strategies for different regions: for overlapping areas, it uses pixel information from integer multiple distances away in already-decoded regions, while for non-overlapping areas, it uses conventional prediction methods. This localized approach optimizes code compression for each region based on its specific characteristics, reducing overall code amount while maintaining prediction accuracy
3Productivity
If conventional prediction methods are used for periodic patterns, then the prediction can be performed, but the prediction accuracy is insufficient
Solution Approach 1:
The patent applies parameter changes by utilizing pixel information at integer multiple distances (N×d, where N is an integer and d is the vector distance) from already-decoded areas. This parameter change allows the prediction to capture periodic patterns more effectively by sampling at multiple scales, thereby improving prediction accuracy for periodic images while maintaining the capability to handle various image types
Data Source
Figure 1~3
Figure 2
Figure 4(1)~4(3)
AI summary
Adopted is a decoder, in which on condition that a prediction block shown by vector information extracted from a data stream, and a decode-target block have an overlap where respective pixels overlay each other, pixel information of an already-decoded portion at a distance of an integer multiple of a vector provided by the vector information from the overlap is made a prediction signal instead of the overlap, and the prediction signal is added to difference image data taken from the data stream to generate reproduction image data. The decoder is adopted for an intra-frame decoder, a local decoder of an encoder, and the like. According to a fundamental rule concerning a repetitive pattern of an image, a pixel at a distance of an integer multiple is a like pixel, and therefore the process of decoding can be performed efficiently.