Segmentation Boundary Smoothing After Image Size Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The boundary of segmented regions becomes conspicuous when a segmentation result after size conversion is applied to the original image, particularly in three-dimensional medical images with varying pixel spacings.
Innovation Solution
An image processing apparatus and method that utilizes a neural network to segment a size-converted image into regions, applies an interpolation calculation to convert class images back to the original image size, and derives the segmentation result based on pixel values to smooth the boundaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the image size is reduced for segmentation processing, then processing speed is improved, but boundary quality deteriorates (jaggies become conspicuous)
Solution Approach 1:
The patent divides the processing into two stages: first segmenting the downsampled image to obtain rough boundaries, then processing each segmented region separately by upsampling with boundary smoothing. This segmentation approach allows different processing strategies for different regions, improving overall boundary quality while maintaining processing efficiency.
Solution Approach 2:
The patent applies local quality improvement by performing boundary smoothing specifically at the boundaries of segmented regions during the upsampling process, rather than uniformly processing the entire image. This targeted approach maintains high boundary quality where it matters most while preserving the computational benefits of downsampling.
2Measurement precision
If the image size is converted to match Pixel Spacing in all directions, then segmentation accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent changes the pixel spacing parameter by converting the image size to match Pixel Spacing in all directions before segmentation. This parameter adjustment ensures uniform sampling density, which improves segmentation accuracy by eliminating artifacts caused by anisotropic pixel spacing.
Solution Approach 2:
The patent performs preliminary image size conversion to match Pixel Spacing before the segmentation process. This preliminary action prepares the image data in advance, ensuring that the segmentation algorithm operates on uniformly spaced data, which simplifies the segmentation process and improves accuracy.
3Productivity
If segmentation is performed on downsampled image and result is applied to original image, then processing efficiency is improved, but boundary smoothness deteriorates
Solution Approach 1:
The patent introduces dynamic boundary smoothing during the upsampling process, where the boundary of each segmented region is adaptively smoothed based on local image characteristics. This dynamic processing maintains the computational efficiency of downsampling while improving boundary smoothness in the final result.
Solution Approach 2:
The patent introduces an intermediate processing step where segmented regions from the downsampled image are used as guides for upsampling the original image. This intermediary approach allows the efficient downsampling segmentation to benefit the high-resolution original image without directly transferring the jagged boundaries.
Data Source
AI summary
A processor is configured to convert a size of a target image to derive a size-converted image, segment the size-converted image into regions of at least one class by using a segmentation model constructed by machine-learning a neural network to derive a plurality of class images in which a pixel value of each pixel represents class-likeness for the at least one class, convert a size of at least one class image into the size of the target image to derive at least one converted class image, and segment the target image based on a pixel value in each pixel of the at least one converted class image.