Misdetection Verification in Image Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for automatic defect detection in digitized image sequences, such as dirt in motion picture films, often result in false negatives and false positives, leading to inefficient and inaccurate identification of defective objects, which can degrade the quality of restored images.
Innovation Solution
A method and apparatus that determine whether a detected defective object is a misdetection by calculating similarity measures between the object's boundary and a replacement pattern's boundary, and between the object's pixels and the replacement pattern's pixels, using acceptance ranges to identify mismatches, thereby distinguishing correct detections from misdetctions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automatic detection algorithms are used to identify defective objects in digitized image sequences, then the productivity of film restoration is improved, but the reliability of detection deteriorates due to false positives and false negatives
Solution Approach 1:
The patent introduces an intermediary verification step between detection and removal that uses similarity comparison with replacement patterns. This mediator process compares the detected object's characteristics against potential replacement regions to verify whether the detection is genuine before proceeding with restoration, thereby reducing false positives while maintaining automated workflow efficiency
Solution Approach 2:
The system implements feedback by using the replacement pattern matching results to confirm or reject detections. The verification process provides feedback about the likelihood of a detection being correct based on how well the detected object's boundary and pixel characteristics match against candidate replacement regions, allowing the system to adjust its confidence in each detection
2Reliability
If manual review is performed to correct misdetections, then the reliability of defect detection is improved, but the productivity of the restoration process deteriorates due to time consumption
Solution Approach 1:
The system performs self-verification by automatically comparing detected objects against replacement patterns from the same image sequence. The algorithm independently validates its own detections by checking whether the detected defective object's boundary and pixel characteristics are inconsistent with any plausible replacement pattern, thereby catching misdetections without requiring manual intervention
Solution Approach 2:
The patent applies preliminary verification before the actual restoration operation. By pre-comparing the detected object against potential replacement patterns and calculating similarity measures, the system identifies and flags likely misdetections before any restoration work is performed, allowing operators to review only the cases that need manual attention
3Reliability
If the boundary similarity measure is used to verify detections, then the reliability of misdetection identification is improved, but the device complexity increases due to additional processing steps
Solution Approach 1:
The verification process is segmented into two independent similarity measures: boundary similarity and pixel similarity. This segmentation allows the system to evaluate different aspects of the detection independently - the shape/contour match separately from the internal texture/pixel match - making the complexity manageable and the measures computationally efficient
Solution Approach 2:
The patent applies different quality criteria to different parts of the comparison. The boundary similarity measure evaluates the contour/shape characteristics locally, while the pixel similarity measure evaluates the internal texture and intensity distribution. This local quality approach allows each measure to focus on specific features relevant to its purpose
Data Source
AI summary
Results of automatic detection of dirt or other non-steady defects in a sequence of digitized image frames can be unreliable. Here, a determination of a detection of a defective object to be replaced by a replacement pattern in a frame of a sequence of image frames as a misdetection is presented that comprises determining a value of a first similarity measure between a boundary of the detected defective object and a boundary of the replacement pattern and determining a value of a second similarity measure between the detected defective object and the replacement pattern. The detection of the defective object is determined as a misdetection if at least one of the following holds true: the value of the first similarity measure is outside of a corresponding first similarity acceptance range and the value of the second similarity measure is inside of a corresponding second similarity acceptance range.