Image Segmentation via Adjacent Segment Joining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation techniques struggle to reliably classify pixels in images, particularly in biological applications like microscopy, as they often neglect spatial relationships between pixels, leading to suboptimal segmentation results, especially when dealing with images of cells and their components like nuclei.
Innovation Solution
A method that segments images based on color values, generates a list of adjacent segment pairs, and selectively joins them according to pre-defined criteria, ensuring the resulting average color difference and size thresholds are met, to improve the accuracy of pixel classification associated with features of interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If pixels are classified according to feature space groupings without considering spatial relationships, then the segmentation process is simpler and faster, but the segmentation accuracy deteriorates because important spatial information is not used
Solution Approach 1:
The patent applies segmentation by dividing the image into multiple regions based on color similarity, then refining these regions by considering spatial relationships. The algorithm segments pixels into color-based groups and then adjusts boundaries based on spatial continuity, resolving the contradiction between simple color-based segmentation and accurate spatial-aware segmentation.
Solution Approach 2:
The patent transitions from two-dimensional color space analysis to three-dimensional analysis by incorporating the spatial dimension. Instead of only considering color features in feature space, the algorithm adds spatial position as a third dimension, allowing pixels to be classified based on both color similarity and spatial proximity, thereby improving accuracy without significantly increasing computational complexity.
2Measurement precision
If spatial relationships between pixels are considered during segmentation, then segmentation accuracy improves, but the algorithm complexity and processing time increase
Solution Approach 1:
The patent uses segmentation to create initial color-based regions and then refines them using spatial information. By breaking down the complex task into two stages (color segmentation followed by spatial refinement), the algorithm achieves high accuracy while keeping the overall complexity manageable through systematic processing.
Solution Approach 2:
The patent performs preliminary color-based segmentation before incorporating spatial relationships. This preliminary action creates an initial segmentation that can be refined later, allowing the algorithm to handle spatial complexity systematically rather than all at once, thereby reducing overall algorithmic complexity while maintaining high accuracy.
3Productivity
If existing segmentation schemes are used on biological images with varying morphological characteristics, then processing is faster, but reliability deteriorates due to inability to handle diverse features reliably
Solution Approach 1:
The patent applies parameter changes by adjusting segmentation criteria based on the specific characteristics of biological images. The algorithm modifies color thresholds, spatial continuity parameters, and region merging criteria to adapt to varying morphological characteristics of cells and tissues, thereby improving reliability across different biological samples while maintaining efficient processing speeds.
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
A method for classifying pixels in an image, e.g. a microscopy image, as being associated with a feature of interest has been described. A color brightfield microscopy image represented by color values for an array of pixels is conventionally obtained. The image is over-segmented based on the color values to provide a plurality of groups of contiguous pixels with related color values, whereby a typical feature of interest will be represented by multiple segments. A list of pairs of segments which are adjacent to one another in the image is generated, and a difference in average color value between the segments comprising each pair is determined. Pairs of adjacent segments are then selectively joined together to form a joined segment to replace the corresponding pair of segments in the list if pre-defined joining criteria are met. The pairs are considered for joining in an order based on their respective differences in average color value. Pixels are then classified as being associated with the feature of interest based on the segments following the selective joining. Segmentation information indicating the pixels classified as being associated with the feature of interest is thus provided.