Image Processing Apparatus Grouping Regions for Filter Operation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The local shared weight filter operation requires a large number of filter kernels, leading to increased memory usage, and existing methods like low rank approximation are not effective in reducing the number of filter kernels, especially when the number of kernels increases.
Innovation Solution
The technique involves grouping weight shared local regions into clusters and using a shared filter kernel for each group, reducing the number of filter kernels needed by performing filter operations using in-group shared filter kernels, which are smaller in size and number compared to local shared filter kernels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If local shared weight filter operation is used to perform filter operation on each region, then filtering accuracy is improved, but the number of filter kernels increases leading to increased memory usage
Solution Approach 1:
The image is divided into multiple local regions, and a separate filter kernel is assigned to each region. This segmentation allows the system to use fewer filter kernels overall while maintaining high filtering accuracy within each local region, thereby resolving the contradiction between filtering accuracy and the number of filter kernels.
2Measurement precision
If the number of filter kernels is increased for local shared weight filter operation, then filtering precision is improved, but processing time increases
Solution Approach 1:
By segmenting the image into local regions and applying dedicated filter kernels to each, the system achieves high filtering precision without the need to process the entire image with multiple kernels sequentially, thereby reducing overall processing time.
Solution Approach 2:
Instead of applying filter kernels to every possible position in the image, the method applies filters to selected local regions only. This partial action approach maintains filtering precision where needed while avoiding redundant computations, thus reducing processing time.
3Measurement precision
If the number of filter kernels is increased for local shared weight filter operation, then filtering accuracy is improved, but power consumption increases
Solution Approach 1:
Segmenting the processing into local regions with dedicated filters reduces the total computational load compared to applying many kernels across the entire image, thereby maintaining filtering accuracy while reducing power consumption.
Solution Approach 2:
By performing filter operations only on selected local regions rather than the entire image, the system reduces the number of computational operations required, which directly lowers power consumption while preserving filtering accuracy in the regions of interest.
Data Source
AI summary
By a first grouping pattern, a plurality of regions for which a filter operation is to be performed in an operation target image are grouped. Then, for each group obtained by the first grouping pattern, a first filter corresponding to the group is used to perform a filter operation on each region belonging to the group. By a second grouping pattern, a plurality of regions for which a filter operation is to be performed in an intermediate image obtained by the filter operation are grouped. Then, for each group obtained by the second grouping pattern, a second filter corresponding to the group is used to perform a filter operation on each region belonging to the group.