Greyscale Edge Detection With Nonlinear Pixel Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional shape-based edge detection algorithms struggle to distinguish between structurally similar shapes in greyscale images, particularly in semiconductor examination, due to inherent linearity, leading to ambiguous edge detection results.
Innovation Solution
A computer-implemented method combining shape-based analysis with greyscale dependent transformation to enhance the visibility of desired edges while diminishing other edges, using regularization coefficients to refine pixel values and improve edge detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional shape-based analysis is applied to greyscale images, then edge detection can be performed, but similar edges with linearly similar pixel values cannot be distinguished
Solution Approach 1:
The patent transforms the greyscale image using a non-linear transformation function that maps pixel values to a new scale, thereby changing the parameter representation of similar edges. This non-linear transformation amplifies the differences between previously indistinguishable edges, enabling their discrimination while maintaining edge detection capability.
Solution Approach 2:
The patent introduces a new dimension by applying a monotonic transformation function that maps the original greyscale values to transformed values in a different numerical space. This dimensional transformation allows edges that were linearly similar to become distinguishable in the transformed space, effectively adding discriminatory power without losing original edge information.
2Reliability
If edge detection is applied to semiconductor specimens, then manufacturing monitoring is enabled, but similar edges produce ambiguous detection results
Solution Approach 1:
By applying non-linear transformation to the greyscale values, the patent changes the parameter scale such that edges with similar original values become distinguishable. This parameter transformation directly addresses the difficulty of detecting and measuring similar edges in semiconductor specimens, thereby improving manufacturing process monitoring reliability.
Solution Approach 2:
The patent replaces the traditional linear edge detection mechanism with a non-linear transformation approach. This substitution transforms the detection mechanism from one that cannot distinguish similar edges to one that can, by fundamentally changing how pixel values are processed and compared.
3Productivity
If conventional edge detection algorithms are used, then processing speed is maintained, but discrimination between linearly similar edges is lost
Solution Approach 1:
The patent applies a non-linear transformation function that changes the parameter representation of pixel values. This transformation can be computed efficiently using pre-computed lookup tables or optimized algorithms, maintaining processing speed while significantly improving edge distinction accuracy by making previously similar edges distinguishable.
Solution Approach 2:
The patent performs a preliminary non-linear transformation of the greyscale image before edge detection. This preliminary action transforms the image data into a form where edges are more distinguishable, and subsequent edge detection operates on this transformed data, achieving both speed and accuracy through the pre-processing step.
Data Source
AI summary
According to the presently disclosed subject matter, to improve edge detection obtained by traditional shape-based analysis and obtain edge detection accurately targeting a certain desired feature (e.g., contours, shape and/or pattern) in the image, greyscale dependent transformation is applied in addition to the shape-based analysis. In this manner the shape-based analysis identifies the shape or pattern of interest, and the greyscale dependent transformation further transforms the shape-based analysis output, such that visibility of pixels that fall within a predetermined pixel value range is increased. By this, ambiguities that result from the inability of the shape-based analysis to discriminate between similar edges up to a linearity, are resolved, and the desired feature (e.g., contour shape and/or pattern) can be identified.