Boundary Padding for Asymmetric Gaussian Blur in OpenCV

Boundary Padding Strategy for Asymmetric Gaussian Blur on Non‑Square Images

0 reputation · 04 Feb 2022, 13:09 UTC

85.3K views0

Goal

Investigate how OpenCV’s GaussianBlur handles boundary padding when an asymmetric kernel (e.g., 5×7) is applied to a non‑square image.

Context

The GaussianBlur function supports various depths (CV_8U, CV_16U, CV_16S, CV_32F, CV_64F) and automatically calculates kernel size when zero is specified. While the implementation uses separable filters for efficiency, the exact padding strategy for asymmetric kernels on rectangular images remains undocumented.

Questions

What padding strategy does OpenCV use internally when the kernel is asymmetric and the image is non‑square?
Does the default border type (e.g., BORDER_DEFAULT) apply uniformly to all sides, or is it adapted per dimension?
Can the padding behavior be overridden or queried programmatically, and if so, what API calls expose this?

Keep exploring this topic

VideoCapture resolution fallback behavior across V4L2 and GStreamer backends

OpenCV's VideoCapture silently falls back to a default resolution when the requested width/height is unsupported, with no error and no cross-platform way to list valid resolutions. How to detect mismatches and discover supported modes reliably?

Strict Integer as Categorical vs Heuristic Inference in OpenCV TrainData: Which Ensures Consistent Feature Types?

How should OpenCV’s TrainData class infer variable types when a column mixes integers and floats? Should it enforce stricter rules, treating integer‑only columns as categorical, or continue with its current heuristic?

gatsby-plugin-image vs gatsby-transformer-sharp for CI builds with bandwidth limits and Sharp dependency concerns

Deciding between gatsby-plugin-image and gatsby-transformer-sharp for CI builds with limited bandwidth and Sharp dependency concerns.

OpenCV TickMeter vs External Profiler for Primitive-Level Bottleneck Attribution

OpenCV's TickMeter gives per-stage wall times but cannot attribute cycles inside a single optimized primitive. The open question is whether external profilers are required for actionable bottleneck identification across varying backend configurations.

Boundary Padding Strategy for Asymmetric Gaussian Blur on Non‑Square Images

Goal

Context

Questions

0 answers

Your perspective belongs here.

0 question comments

Good answers start here.