Enhance

Targets:

image

Image Types:uint8, float32

Mild Pillow-inspired local enhancement (edge or detail) blended with the original via alpha. Use for subtle contour or detail boost milder than Sharpen.

A native Albumentations implementation of the Pillow EDGE_ENHANCE / EDGE_ENHANCE_MORE and DETAIL filter family. The enhanced image is computed via a small 3x3 convolution and then blended with the original:

output = (1 - alpha) * image + alpha * enhanced_image

Equivalently, the convolution is applied with the precomputed kernel K(alpha) = (1 - alpha) * I + alpha * E where E is the mode-specific kernel. Because each E sums to 1, K(alpha) also sums to 1 and brightness is preserved.

For mode="edge", alpha=1 reproduces Pillow's EDGE_ENHANCE and alpha=2 reproduces EDGE_ENHANCE_MORE. For mode="detail", alpha=1 reproduces Pillow's DETAIL. Values of alpha between 0 and 1 give milder presets; values above 1 overshoot for stronger effects.

Arguments

mode

edge | detail

edge

Which native enhancement operator to use.

"edge": crispens contours and boundaries (Pillow EDGE_ENHANCE family).
"detail": mild local detail / fine-structure boost (Pillow DETAIL). Default: "edge".

alpha_range

tuple[float, float]

[0.5,1]

Range from which the blend strength alpha is sampled uniformly per call. alpha=0 is no-op, alpha=1 is the full Pillow preset, alpha>1 overshoots into a stronger variant. Must be non-decreasing with non-negative values. Default: (0.5, 1.0).

p

float

0.5

Probability of applying the transform. Default: 0.5.

Examples

>>> import numpy as np
>>> import albumentations as A
>>> image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)
>>>
>>> # Edge enhancement, mild to Pillow-EDGE_ENHANCE strength
>>> transform = A.Compose([A.Enhance(mode="edge", alpha_range=(0.5, 1.0), p=1.0)])
>>> result = transform(image=image)["image"]
>>>
>>> # Stronger edge variant (alpha=2 matches Pillow EDGE_ENHANCE_MORE)
>>> transform = A.Compose([A.Enhance(mode="edge", alpha_range=(1.0, 2.0), p=1.0)])
>>> result = transform(image=image)["image"]
>>>
>>> # Subtle detail / fine-structure boost
>>> transform = A.Compose([A.Enhance(mode="detail", alpha_range=(0.5, 1.0), p=1.0)])
>>> result = transform(image=image)["image"]

Notes

Edge handling follows the rest of Albumentations (cv2 BORDER_REFLECT_101), which differs slightly from Pillow's clamping at borders.
For uint8 inputs the output saturates to [0, 255]; for float32 it is clipped to [0, 1] by convolve.

See Also

Sharpen: Broader high-frequency sharpening (kernel-Laplacian or unsharp-mask). Use when you need a continuous, configurable sharpening primitive rather than a compact preset family.
UnsharpMask: Classic unsharp-mask sharpening with explicit blur radius.
Emboss: Directional edge highlight for stylization rather than enhancement.

References

Pillow ImageFilter (EDGE_ENHANCE, EDGE_ENHANCE_MORE, DETAIL)https://pillow.readthedocs.io/en/stable/reference/ImageFilter.html

>>> import numpy as np >>> import albumentations as A >>> image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8) >>> >>> # Edge enhancement, mild to Pillow-EDGE_ENHANCE strength >>> transform = A.Compose([A.Enhance(mode="edge", alpha_range=(0.5, 1.0), p=1.0)]) >>> result = transform(image=image)["image"] >>> >>> # Stronger edge variant (alpha=2 matches Pillow EDGE_ENHANCE_MORE) >>> transform = A.Compose([A.Enhance(mode="edge", alpha_range=(1.0, 2.0), p=1.0)]) >>> result = transform(image=image)["image"] >>> >>> # Subtle detail / fine-structure boost >>> transform = A.Compose([A.Enhance(mode="detail", alpha_range=(0.5, 1.0), p=1.0)]) >>> result = transform(image=image)["image"]