Brightness Contrast

Overview

The Brightness Contrast sample demonstrates GPU-accelerated brightness and contrast adjustment using CV-CUDA’s brightness_contrast operator. The operator applies a per-image affine transform to every pixel:

output = brightness * (contrast * (input − contrast_center) + contrast_center) + brightness_shift

where brightness, contrast, brightness_shift, and contrast_center are per-image scalar tensors. This formulation separates multiplicative brightness from the contrast pivot, giving fine-grained control over image appearance.

Usage

Basic Usage

Apply the default brightening/contrast boost to the bundled test image:

python3 brightness_contrast.py

Custom Input

Supply your own image:

python3 brightness_contrast.py -i input.jpg -o output.jpg

Command-Line Arguments

Argument

Short Form

Default

Description

--input

-i

tabby_tiger_cat.jpg

Input image file path

--output

-o

cvcuda/.cache/cat_brightness_contrast.jpg

Output image file path

Implementation

Parameter Setup

# The brightness_contrast operator expects per-image scalar parameters
# supplied as 1-D tensors with layout "N" (one element per image in the batch).
# brightness: multiplicative gain applied to the pixel values (>1 brightens).
# contrast:   multiplier around contrast_center (<1 compresses, >1 expands).
# brightness_shift: additive offset added after brightness scaling.
# contrast_center: the pivot value around which contrast is computed (default 0.0).
brightness_host = np.array([1.5], dtype=np.float32)
contrast_host = np.array([1.4], dtype=np.float32)
brightness_shift_host = np.array([20.0], dtype=np.float32)
contrast_center_host = np.array([127.0], dtype=np.float32)

brightness = cvcuda.Tensor((1,), dtype=cvcuda.Type.F32, layout="N")
contrast = cvcuda.Tensor((1,), dtype=cvcuda.Type.F32, layout="N")
brightness_shift = cvcuda.Tensor((1,), dtype=cvcuda.Type.F32, layout="N")
contrast_center = cvcuda.Tensor((1,), dtype=cvcuda.Type.F32, layout="N")

cuda_memcpy_h2d(brightness_host, brightness.cuda())
cuda_memcpy_h2d(contrast_host, contrast.cuda())
cuda_memcpy_h2d(brightness_shift_host, brightness_shift.cuda())
cuda_memcpy_h2d(contrast_center_host, contrast_center.cuda())

Operator Call

# The brightness_contrast operator expects per-image scalar parameters
# supplied as 1-D tensors with layout "N" (one element per image in the batch).
# brightness: multiplicative gain applied to the pixel values (>1 brightens).
# contrast:   multiplier around contrast_center (<1 compresses, >1 expands).
# brightness_shift: additive offset added after brightness scaling.
# contrast_center: the pivot value around which contrast is computed (default 0.0).
brightness_host = np.array([1.5], dtype=np.float32)
contrast_host = np.array([1.4], dtype=np.float32)
brightness_shift_host = np.array([20.0], dtype=np.float32)
contrast_center_host = np.array([127.0], dtype=np.float32)

brightness = cvcuda.Tensor((1,), dtype=cvcuda.Type.F32, layout="N")
contrast = cvcuda.Tensor((1,), dtype=cvcuda.Type.F32, layout="N")
brightness_shift = cvcuda.Tensor((1,), dtype=cvcuda.Type.F32, layout="N")
contrast_center = cvcuda.Tensor((1,), dtype=cvcuda.Type.F32, layout="N")

cuda_memcpy_h2d(brightness_host, brightness.cuda())
cuda_memcpy_h2d(contrast_host, contrast.cuda())
cuda_memcpy_h2d(brightness_shift_host, brightness_shift.cuda())
cuda_memcpy_h2d(contrast_center_host, contrast_center.cuda())

Key points:

  1. Parameter tensors — each of brightness, contrast, brightness_shift, and contrast_center is a 1-D "N"-layout float32 tensor with one element per image.

  2. All parameters are optional — you may pass any subset; omitted parameters default to identity values (brightness=1, contrast=1, brightness_shift=0, contrast_center=0).

  3. Dtype preserved — the output tensor has the same dtype and layout as the input, so no post-processing conversion is needed for uint8 HWC images.

  4. contrast_center pivot — setting contrast_center to 127.0 for uint8 data places the pivot at mid-gray, which preserves overall luminance while expanding tonal range.

Expected Output

../../_images/tabby_tiger_cat.jpg

Original Input Image
../../_images/cat_brightness_contrast.jpg

Output: Brightness × 1.5, Contrast × 1.4

CV-CUDA Operators Used

Operator

Purpose

cvcuda.brightness_contrast()

Per-image affine pixel transform for brightness and contrast adjustment

Common Utilities Used

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

  • write_image() - Save adjusted image

  • cuda_memcpy_h2d - Upload per-image scalar parameters from host to device

See Also