Template Matching Apparatus Using Valley Enhancement for Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional template matching techniques fail to accurately detect objects when the object region in the input image is significantly transformed compared to the template image or when noise is present, leading to incorrect matching and threshold setting challenges, especially in applications like micromachining and semiconductor inspection.
Innovation Solution
A template matching technique that transforms the input image to match the object region of the template image, using valley enhancement and distance mapping to shape non-background regions, thereby minimizing the impact of noise and positional shifts, and computes a normalized correlation value for accurate object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional template matching is used with transformed input images, then matching operation speed is maintained, but detection accuracy deteriorates when object region transformation exceeds template image transformation capability
Solution Approach 1:
The patent segments the image processing into distinct modules: valley enhancement processing for noise reduction, distance mapping processing for transformation compensation, and correlation coefficient calculation for matching. This segmentation allows each module to handle specific aspects of the complexity independently, improving overall detection accuracy without overwhelming computational burden.
Solution Approach 2:
The patent applies preliminary transformations to the input image before the main matching operation. Specifically, valley enhancement is performed first to reduce noise, then distance mapping is applied to compensate for transformations. These preliminary actions prepare the image data in advance, ensuring that when the correlation coefficient calculation occurs, the data is already optimized for accurate matching, thereby improving detection accuracy without increasing the complexity of the main matching algorithm.
2Reliability
If noise is present in the input image, then real-world applicability is improved, but matching reliability deteriorates due to incorrect correlation values
Solution Approach 1:
The patent converts the harmful effect of noise into a beneficial process by applying valley enhancement. This technique specifically targets and reduces noise in the input image, transforming the noisy data into cleaner data that produces more reliable correlation values. By doing so, the patent maintains real-world applicability (dealing with noisy images) while simultaneously improving matching reliability through systematic noise reduction.
3Measurement precision
If threshold value is set high to ensure accuracy, then detection precision is improved, but false negative rate increases due to transformed regions not matching template
Solution Approach 1:
The patent changes the parameters of the input image through valley enhancement and distance mapping transformations. These parameter changes modify the image characteristics to better match the template, thereby increasing the correlation coefficient values. As a result, even transformed regions achieve sufficient match scores to pass the threshold, improving detection precision without requiring overly high threshold values that would cause false negatives.
Data Source
AI summary
A matching degree computing apparatus is provided for comparing an input image and an object template image and computing a matching degree between an input image and an object template image based on the compared result. The computing apparatus includes a transforming unit for transforming the input image so as to be matched to the template object region and a computing unit for computing a matching degree between the transformed input image and the template image. The transforming unit provides a shaping unit for shaping a non-background region to the form of the template object region in the object corresponding region of the input image and a processing unit for arranging the non-background region contacting with the template object corresponding region so that the non-background region has no substantial impact on the matching degree in the object non-corresponding region of the input image.