Laplacian

Overview

The Laplacian sample demonstrates second-order edge detection using CV-CUDA’s GPU-accelerated Laplacian operator. The Laplacian highlights regions of rapid intensity change, making it a classic tool for detecting edges and fine structures in images. After applying the operator the sample stretches the response histogram to the full uint8 range so the edge map is immediately viewable as a JPEG.

Usage

Basic Usage

Apply the Laplacian operator with the default settings (ksize=3, scale=1.0):

python3 laplacian.py -i input.jpg

Custom Output Path

Specify a custom output file path:

python3 laplacian.py -i input.jpg -o cat_laplacian.jpg

Command-Line Arguments

Argument

Short Form

Default

Description

--input

-i

tabby_tiger_cat.jpg

Input image file path

--output

-o

cvcuda/.cache/cat_laplacian.jpg

Output image file path

Implementation

Laplacian Edge Detection

# Apply the Laplacian operator.
# ksize=3 selects the 3×3 discrete Laplacian aperture; scale=1.0 leaves
# the computed values unchanged before the output is saturated back to uint8.
# REPLICATE border avoids zero-valued artifacts at the image boundary.
output_image: cvcuda.Tensor = cvcuda.laplacian(
    input_image,
    ksize=3,
    scale=1.0,
    border=cvcuda.Border.REPLICATE,
)

# The raw Laplacian response occupies only a small fraction of [0, 255].
# Stretch the histogram so the edges are clearly visible in the saved JPEG.
output_image = normalize_to_uint8(output_image)

write_image(output_image, args.output)

Key points:

  1. Kernel size: ksize=3 selects the 3×3 discrete Laplacian aperture. The only other supported value is ksize=1, which uses a simpler cross-shaped kernel.

  2. Scale factor: scale=1.0 applies a uniform multiplier to the computed Laplacian values before the result is saturated back to the input dtype. Increasing the scale amplifies weaker edges.

  3. Border handling: cvcuda.Border.REPLICATE repeats the edge pixels outward, avoiding the zero-filled boundary artifacts that CONSTANT mode would introduce.

  4. Dtype preservation: The operator returns a tensor with the same dtype and layout as the input (uint8 HWC here), so no format conversion is required.

  5. Histogram stretching: The raw Laplacian response is typically concentrated in a narrow value range. A host-side min/max stretch makes the edge map clearly visible in the saved image without changing the operator’s output semantics.

Expected Output

The output shows the Laplacian edge response of the input image, normalized for visibility:

../../_images/tabby_tiger_cat.jpg

Original Input Image
../../_images/cat_laplacian.jpg

Output: Laplacian Edge Response

CV-CUDA Operators Used

Operator

Purpose

cvcuda.laplacian()

Apply second-order Laplacian edge-detection filter to the input image

Common Utilities Used

  • read_image() - Load image as CV-CUDA tensor

  • write_image() - Save the edge-response image

  • cuda_memcpy_d2h / cuda_memcpy_h2d - Transfer tensor data to/from host for histogram stretching

See Also