Wavelet Image Decomposition Scheduling for Memory Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing architectures for two-dimensional discrete wavelet transforms require significant memory resources, leading to increased component size and complexity, which limits their practical application in embedded systems.
Innovation Solution
A device with a scheduling mechanism that optimizes processing tasks across multiple levels of decomposition, using a routing block to configure processing modules based on the current decomposition level, allowing a single line to be associated with a single level of decomposition, thereby reducing memory requirements and implementation complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing architectures for two-dimensional discrete wavelet transforms are used, then multi-level decomposition functionality is achieved, but memory resources increase significantly
Solution Approach 1:
The patent segments the wavelet transform process into distinct processing stages (horizontal transform, vertical transform, and reconstruction stages), with each stage handling specific decomposition levels. This segmentation allows memory to be reused across stages rather than requiring separate memory for each decomposition level, thereby reducing overall memory requirements while maintaining full multi-level decomposition functionality.
Solution Approach 2:
The patent transitions from a spatial memory organization to a temporal reuse strategy by processing different decomposition levels in sequence rather than simultaneously. The scheduling mechanism orchestrates the execution of processing tasks across time, allowing the same memory resources to serve multiple decomposition levels at different time intervals, thus reducing memory quantity while preserving functional capability.
2Reliability
If existing architectures for two-dimensional discrete wavelet transforms are used, then decomposition capability is maintained, but component size and complexity increase
Solution Approach 1:
The patent designs processing modules that perform multiple functions: the same horizontal transform module serves both level-1 and higher-level decompositions, and the same vertical transform module is reused across all decomposition levels. This multi-functionality reduces the number of separate components needed, thereby reducing device complexity and size while maintaining complete decomposition capability across all levels.
Solution Approach 2:
The patent introduces a dynamic scheduling mechanism that adaptively configures the processing pipeline based on the current decomposition level being processed. The scheduler dynamically activates or deactivates specific processing paths and adjusts data flow routing, allowing a single configurable architecture to replace multiple fixed architectures, thus reducing overall system complexity while preserving full decomposition functionality.
Data Source
Figure 1a~3
Figure 4~5
Figure 6~7
AI summary
The invention relates to a device for decomposing an image into at least three levels using a wavelet transform, and comprising a first bloc (500) for implementing a first level of decomposition and a second bloc (501) for implementing the higher levels of decomposition by executing a series of processing tasks. Said tasks are scheduled in time using a series of lines, a routing bloc (502) having the function of configuring the second bloc (501) when the level of decomposition associated with the processing task being executed changes relative to the level of decomposition associated with the processing task previously executed. The processing tasks are scheduled such that a given line is associated with only one level of decomposition.