EDIUpsizer: The Mechanics of Advanced Edge-Directed Video Upscaling
EDIUpsizer is a classical video resampling filter developed for AviSynth that upscales digital images and video frames by a factor of two. Unlike standard bilinear or bicubic algorithms that often yield blurry results, this specialized plugin reverses upsampling artifacts by using New Edge-Directed Interpolation (NEDI).
This technical overview explores how the filter balances structural sharpness with artifact control in classic digital video processing. The Architecture: How EDIUpsizer Works
Standard video scaling methods look at neighboring pixels and average their values. While fast, this approach ignores the visual flow of edges, making diagonal lines look jagged (aliased) or blurry.
EDIUpsizer fixes this by calculating local covariances in the lower-resolution source frame. It then applies these data points to adapt the high-resolution output. This allows the filter to predict where an edge continues and accurately fill in the missing pixels.
[Low-Resolution Source] │ ▼ [Local Covariance Analysis] ──► Measures edge orientation │ ▼ [Adaptive Window Sizing] ────► Prevents artifacts in high-detail areas │ ▼ [Edge-Directed Interpolation] ──► Sharp 2x Horizontal & Vertical Scale Key Features and Artifact Control
While pure NEDI interpolation improves edge clarity, it frequently generates unwanted visual noise or “artifacts” in fine, complex textures. The developer of EDIUpsizer added several key modifications to the core algorithm to counteract this drawback:
Adaptive Window Sizing: The filter dynamically adjusts its sampling area based on the complexity of the image texture.
Condition Number Testing: This statistical check evaluates the reliability of local pixel data before processing, keeping the upscale clean.
Capping Constraints: This feature restricts output limits to keep the generated pixels stable and natural-looking.
User Adjustments: All of these individual safety constraints can be turned on, off, or fine-tuned via the script interface. Performance: The Speed Trade-off
The high structural accuracy of this filter comes at a cost. EDIUpsizer is computationally heavy and slow. Because it analyzes math models for every pixel block, it is best suited for offline rendering rather than real-time playback.
To address this performance bottleneck, the developer released a sibling plugin called FastEDIUpsizer. It cuts processing times using a simplified architecture:
It uses a fixed 8×8 pixel window size instead of dynamic scaling.
It restricts NEDI analysis to the luma (brightness) plane, where human eyes see sharpness.
It falls back to cheaper bicubic or bilinear math for the color data. Summary: Comparison of Methods Bilinear / Bicubic EDIUpsizer Edge Sharpness Blurry / Soft Sharp Artifact Control Low (Noisy) High Processing Speed Very Slow Best Use Case Real-time video Line-art scaling High-quality video preservation
If you are working on a video restoration project, let me know:
What type of footage you are processing (such as real-world video, animation, or pixel art). Your target frame rate and rendering time limits. External plugins old – Avisynth wiki
Leave a Reply