Image Processor Gradient Scanning for Amorphous Cell Area Specification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional cell recognition processes, such as those using circular models or the Watershed algorithm, fail to accurately identify areas in images with non-circular shapes or amorphous cell images, leading to incorrect center extraction and borderline recognition, especially when cells are close to each other or have varying luminance gradients.
Innovation Solution
An image processor with a gradient information calculating unit, extreme value coordinate acquiring unit, and area specifying unit that scans pixels based on gradient information to identify extreme value pixels and label areas, allowing for precise area specification in grayscale images, regardless of cell shape or luminance variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional cell recognition processes using circular models or Watershed algorithm are used, then the processing method is simple and well-defined, but the area specification accuracy deteriorates for non-circular shapes and amorphous cell images
Solution Approach 1:
The patent changes the fundamental parameter used for area specification from model-based assumptions (circular shape, luminance gradients) to extreme value coordinates derived directly from pixel data. By identifying pixels with extreme values (minimum or maximum) and using their coordinates to define area boundaries, the method adapts to any cell shape without requiring shape assumptions, thereby improving both accuracy and versatility
Solution Approach 2:
Instead of approaching the problem by assuming a cell shape model and extracting features from it (conventional approach), the patent inverts the approach by starting with the pixel data itself, identifying extreme value pixels, and deriving the area specification from those extreme points. This inversion allows the method to work with any shape rather than forcing data into a predefined model
2Measurement precision
If circular model is used for center extraction, then the extraction process is straightforward, but the center extraction accuracy deteriorates when cells are close to each other or have non-circular shapes
Solution Approach 1:
The patent extracts only the essential information needed for center identification - the coordinates of extreme value pixels - without requiring full model fitting or complex calculations. By taking out just the extreme point coordinates from the pixel data and using these to define area boundaries, the method achieves accurate center extraction for various shapes while maintaining relatively simple processing
Solution Approach 2:
The patent creates a simplified representation of the cell area by copying only the extreme value pixel coordinates rather than working with the entire complex image data or fitted model parameters. This copied coordinate information is sufficient to define the area boundaries and identify centers, reducing computational complexity while preserving accuracy
Data Source
AI summary
An image processor that can suitably specify a predetermined area of a grayscale image is provided. An extreme value coordinate acquiring unit performs scanning on at least one of a plurality of pixels including a target pixel from the target pixel in accordance with gradient information corresponding to a change in a pixel value, and acquires a coordinate of the scanned pixel corresponding to an extreme value of the pixel value as a coordinate of an extreme value pixel corresponding to the target pixel. An area specifying unit labels each target pixel with an area identification mark used for specifying an area to which the each pixel belongs. The area specifying unit 103 labels each of different pixels corresponding to coordinates of extreme value pixels located close to each other within a particular range with an area identification mark indicating an identity of the area to which the pixels belong. Accordingly, the area to which each pixel belongs is able to be specified based on the area identification mark.