Blur Detection Using Local Sharpness Map
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting blur in video image sequences are overly sensitive to noise and fail to distinguish between artistic blurriness and processing degradation, making it difficult to assess the sharpness of HD video content that may have been up-converted from SD material, and they do not effectively differentiate between blur caused by compression artifacts and actual image characteristics.
Innovation Solution
A single-ended blur detection method using two sets of edge filters, one for 'fast' and one for 'slow' edges, which produce weighted sums and decimated arrays to calculate a localized maximum sharpness value, reducing noise sensitivity and identifying areas of blur or sharpness through a sharpness overlay map, and optionally disabling horizontal and vertical filters to detect blockiness artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional blur detection methods are used, then blur measurement is provided, but the measurement is overly sensitive to noise which reduces the assigned blur value even though the image still appears soft or blurry
Solution Approach 1:
The patent segments the image processing into distinct filter stages: fast edge detection filters, slow edge detection filters, and differential processing. By dividing the blur detection process into separate filtering operations with different temporal characteristics, the method can distinguish between noise (fast variations) and actual blur (slow variations), thereby reducing noise sensitivity while maintaining measurement accuracy
Solution Approach 2:
The patent introduces differential processing as an intermediary step between fast and slow edge detection. The differential operation (fast edge output minus slow edge output) acts as a mediator that cancels out noise components while preserving the blur measurement signal, effectively mitigating the harmful effect of noise on blur detection accuracy
2Measurement precision
If blur detection is performed on HD video content, then sharpness assessment is provided, but it is difficult to distinguish between artistic blurriness and processing degradation
Solution Approach 1:
The patent employs dynamic filter selection where the system adapts between fast and slow edge detection based on the temporal characteristics of the video content. By dynamically adjusting which filter responses are weighted more heavily, the method can distinguish between static artistic blur (detected by slow filters) and processing degradation (detected by fast filters), enabling accurate sharpness assessment across different content types
3Reliability
If compression artifacts are present in video, then loss of detail occurs, but it is difficult to differentiate between blur caused by compression and actual image characteristics
Solution Approach 1:
The patent applies preliminary filtering operations (fast and slow edge detection) before performing the final differential blur measurement. These preliminary filter stages pre-process the image data to separate compression artifact patterns from actual blur patterns, allowing the subsequent differential operation to reliably distinguish between the two and maintain accurate blur detection despite the presence of compression artifacts
Data Source
Figure 1a~1b
Figure 2a
Figure 2b
AI summary
A single-ended blur detection probe and method with a local sharpness map for analyzing a video image sequence uses two sets of edge filters, one for "fast edges" and the other for "slow edges." Each set of edge filters includes a horizontal bandpass filter, a vertical bandpass filter and a pair of orthogonal diagonal filters where the frequency response of the fast edge filters overlap the frequency response of the slow edge filters. The video image sequence is input to each filter of each set, and the output absolute values are combined with weighting factors to produce a slow edge weighted sum array and a fast edge weighted sum array. The respective weighted sum arrays are then decimated to produce a slow edge decimated array and a fast edge decimated array. The ratio of the maximum difference value between the decimated arrays and the maximum value from the fast edge decimated array, weighted by an appropriate factor, produces a localized maximum sharpness value, the log of which produces a dimensionless blur value.