Implementation:Kornia Kornia Unsharp Mask

Knowledge Sources	Kornia
Domains	Vision, Image_Filtering
Last Updated	2026-02-09 15:00 GMT

Overview

Sharpens images using the unsharp masking technique, which enhances edges by computing the difference between the original image and its Gaussian-blurred version.

Description

This module is part of the Kornia library's filters subpackage. It provides the unsharp_mask function and UnsharpMask nn.Module class for image sharpening. The unsharp masking technique works by subtracting a blurred version of the image from the original and adding the result back, effectively amplifying high-frequency details (edges). The implementation uses the formula: out = 2 * image - gaussian_blur2d(image), which is efficiently computed via torch.lerp(data_blur, input, weight=2.0). The Gaussian blur is applied using gaussian_blur2d from the same filters subpackage.

Usage

Import this function or class when you need to enhance image sharpness, increase edge contrast, or counteract blurring in images, such as for display enhancement, preprocessing before feature detection, or post-processing after upsampling.

Code Reference

Source Location

• Repository: Kornia
• File: kornia/filters/unsharp.py
• Lines: 1-100

Signature

def unsharp_mask(
    input: torch.Tensor,
    kernel_size: tuple[int, int] | int,
    sigma: tuple[float, float] | torch.Tensor,
    border_type: str = "reflect",
) -> torch.Tensor: ...

class UnsharpMask(nn.Module):
    def __init__(self, kernel_size: tuple[int, int] | int,
                 sigma: tuple[float, float] | torch.Tensor,
                 border_type: str = "reflect") -> None: ...
    def forward(self, input: torch.Tensor) -> torch.Tensor: ...

Import

from kornia.filters import unsharp_mask, UnsharpMask

I/O Contract

Inputs

Name	Type	Required	Description
input	torch.Tensor (B, C, H, W)	Yes	The input image tensor to sharpen.
kernel_size	tuple[int, int] or int	Yes	Size of the Gaussian blur kernel used for the unsharp mask.
sigma	tuple[float, float] or torch.Tensor	Yes	Standard deviation of the Gaussian kernel.
border_type	str	No (default "reflect")	Padding mode: constant, reflect, replicate, or circular.

Outputs

Name	Type	Description
output	torch.Tensor (B, C, H, W)	The sharpened image, same shape as input.

Usage Examples

import torch
from kornia.filters import unsharp_mask, UnsharpMask

# Functional unsharp mask
input = torch.rand(2, 4, 5, 5)
output = unsharp_mask(input, (3, 3), (1.5, 1.5))
print(output.shape)  # torch.Size([2, 4, 5, 5])

# Module-based unsharp mask
sharpen = UnsharpMask((3, 3), (1.5, 1.5))
output = sharpen(input)
print(output.shape)  # torch.Size([2, 4, 5, 5])

# Stronger sharpening with larger kernel and sigma
output_strong = unsharp_mask(input, (7, 7), (3.0, 3.0))

Related Pages

• Principle:Kornia_Kornia_Unsharp_Masking