Spectral Power Signature for Tiling Detection in Video
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video compression technologies fail to accurately detect the severity of tiling or blockiness in decoded compressed video signals, especially when the tiling pattern is not aligned with pixel 0,0 or has been affected by image resizing, leading to undetected artifacts.
Innovation Solution
The method employs one-dimensional vectors at block edges to create a spectral power signature by edge-enhancing the baseband image component, summing pixel values, and analyzing power at specific frequencies to produce dimensionless tiling or blockiness values, independent of phase or alignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional AC energy ratio comparison method is used to detect tiling, then detection is simple for pixel 0,0 aligned tiles, but tiling detection fails for non-aligned tiles (motion-compensated frames, resized images)
Solution Approach 1:
The patent transforms the 2D spatial domain tiling detection problem into a 1D frequency domain analysis by projecting horizontal and vertical edge differences onto spectral axes. This dimensional transformation enables detection of tiling patterns regardless of their spatial alignment, as frequency domain representation is invariant to spatial shifts. The spectral power signature approach converts position-dependent spatial patterns into position-independent frequency characteristics.
Solution Approach 2:
The patent replaces the mechanical/spatial comparison method (AC energy ratio between neighboring blocks) with a spectral analysis method. Instead of physically comparing block energies in the spatial domain, the invention uses Fourier transform-based spectral power signature analysis, substituting mechanical spatial operations with frequency domain mathematical transformations that are inherently alignment-invariant.
2Productivity
If lossy compression is applied to reduce data bit-rate, then transmission efficiency improves, but visible tiling artifacts increase in the decoded video
Solution Approach 1:
The patent replaces subjective visual assessment of tiling artifacts with an objective spectral power signature measurement system. The frequency domain analysis quantifies tiling severity by measuring power at specific spatial frequencies characteristic of block boundaries, providing an objective metric that correlates with perceived artifact severity without requiring human visual evaluation.
Solution Approach 2:
The patent changes the measurement parameter from spatial domain AC energy ratios to frequency domain spectral power distribution. By analyzing the spectral power signature at frequencies corresponding to typical block sizes (e.g., 8x8, 16x16 blocks), the system detects tiling artifacts as distinct frequency peaks, enabling quantitative assessment of compression-induced artifacts.
3Adaptability or versatility
If image resizing is performed (e.g., 1080i to 720p conversion), then adaptability to different display formats improves, but tiling patterns become misaligned and undetectable by traditional methods
Solution Approach 1:
The patent transforms the spatial alignment problem into a frequency domain problem where tiling patterns manifest as spectral power peaks regardless of spatial position. The Fourier transform converts position-dependent spatial block patterns into position-independent frequency signatures, allowing detection of tiling artifacts in resized images where the block patterns no longer align with the original pixel grid.
Data Source
AI summary
Tiling or blockiness detection based on spectral power signature uses one-dimensional vectors at block edges to find a spectral signature created by the tiling or blockiness in an image. A baseband component of the image, such as luminance, is edge enhanced, and then the pixel values along each horizontal line are summed to form a one-dimensional column vector of summed edge values for the image. The power of the column vector and the power of selected frequency components within the column vector are determined. The powers are then combined and converted to dimensionless tiling or blockiness values relative to each of the selected frequencies.