Image Processing Device Using Multiple Low-Pass Filters for Edge Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing methods using epsilon filters face challenges in maintaining edge accuracy while reducing processing load and increasing speed, often resulting in blurred edges or vertical outlines when lowering image resolution or applying sequential one-dimensional filters.
Innovation Solution
An image processing device employs multiple low-pass filters with different level value ranges to extract and smooth pixels within specific ranges, using synthesis processing units to calculate and interpolate level values, thereby maintaining edge accuracy and blurring non-edge portions without conditional branch processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If a low pass filter is used to generate a smoothed image, then the portion other than the edge is blurred, but the edge portion is also blurred and cannot be maintained
Solution Approach 1:
The patent applies different processing characteristics to different regions of the image by using multiple low pass filters with different level value ranges. Each filter processes pixels within its specific level range, allowing the image to have different smoothing characteristics in different regions - edges are preserved while non-edge portions are blurred according to their respective level values.
Solution Approach 2:
The patent segments the image processing into multiple parallel low pass filter operations, each handling a specific level value range. This segmentation allows simultaneous processing of different image regions with appropriate smoothing strength, preventing edge blurring while achieving the desired blur in non-edge areas.
2Manufacturing precision
If an epsilon filter is used to maintain edge accuracy, then the edge portion is accurately maintained, but the processing load is heavy due to conditional branch processes
Solution Approach 1:
The patent extracts the conditional branching logic from the processing flow by pre-organizing pixels into multiple low pass filters based on their level values. This extraction eliminates the need for runtime conditional branches, as each pixel is automatically routed to the appropriate filter based on its level value range, significantly reducing processing load while maintaining edge accuracy.
Solution Approach 2:
The patent changes the processing parameter from a single uniform filter approach to multiple filters with different level value ranges. This parameter change allows the system to process different level value ranges in parallel without conditional branches, improving processing speed while maintaining the ability to preserve edges.
3Productivity
If image resolution is lowered to reduce processing load, then the processing speed increases, but edge information may be lost and edge accuracy deteriorates
Solution Approach 1:
The patent adds a new dimension to image processing by introducing multiple level value ranges as an additional processing dimension. Instead of processing all pixels uniformly or reducing resolution, the system processes pixels in parallel across different level value range dimensions, maintaining edge information while achieving high processing speed through this dimensional expansion.
4Device complexity
If sequential one-dimensional filters are applied to reduce processing complexity, then the device complexity is reduced, but vertical outlines are generated and image quality deteriorates
Solution Approach 1:
The patent merges multiple low pass filters with different level value ranges into a unified processing system that operates in parallel. This merging combines the simplicity of one-dimensional filter structures with the quality preservation of multi-range processing, eliminating vertical outlines while maintaining low device complexity through the use of standard filter components.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
LPFs (1121-1 through 1121-n) each set with different ranges of level values, sequentially use pixels in an input image (131) as a pixel of interest, extract a pixel whose level value of pixels including the pixel of interest in a filter size within a range of level values set therein, and generate level value limit smoothed images (132-1 through 132-n) that use an average value of level values of the extracted pixel as a level value of the pixel of interest. A synthesis processing unit (1122) sequentially uses pixels in the input image (131) as a processing object pixel, selects the level value limit smoothed images (132-1 through 132-n) generated by the LPFs (1121-1 through 1121-n) set with a range of level values including a level value of the processing object pixel, and generates a smoothed image (133) that uses a level value of a pixel corresponding to a position of the processing object pixel in the selected level value limit smoothed images (132-1 through 132-n) as the level value of the processing object pixel.