Crate trueno_image 

GPU image processing primitives.

Contract: image-conv2d-v1.yaml

Provides convolution, Gaussian blur, Sobel edge detection, and Canny edge detection with provable properties.

Example

use trueno_image::{conv2d, BorderMode};

// Identity convolution (delta kernel)
let image = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0_f32];
let delta = [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0_f32];
let out = conv2d(&image, 3, 3, &delta, 3, 3, BorderMode::Zero).unwrap();
assert!((out[4] - 5.0).abs() < 1e-6); // Center pixel preserved

Structs

ImageBuf

Structured image buffer with metadata.

Enums

BorderMode

Border handling mode for convolution.

DType

Pixel data type.

ImageError

Errors from image operations.

Interpolation

Interpolation method for resize.

Traits

ImageOps

Image operations trait for ImageBuf method dispatch.

Functions

canny

Canny edge detection.

canny_rgb

Multi-channel Canny edge detection (NPP 3-channel parity).

closing

Morphological closing: dilate then erode.

connected_components

Connected component labeling using union-find (4-connectivity).

conv2d

2D convolution with same-padding.

cumulative_histogram

Compute cumulative histogram (CDF).

dilate

Dilate: output pixel = max of neighborhood defined by structuring element.

equalize

Histogram equalization: remap image to uniform histogram.

erode

Erode: output pixel = min of neighborhood defined by structuring element.

gaussian_blur

Gaussian blur using separable convolution.

gradient_magnitude

Gradient magnitude from Sobel output.

histogram

Compute histogram of a grayscale image with values in [0, 1].

hsv_to_rgb

Convert HSV to RGB color space.

opening

Morphological opening: erode then dilate.

resize

Resize a grayscale image.

rgb_to_gray

Convert RGB image to grayscale using ITU-R BT.601 weights.

rgb_to_hsv

Convert RGB to HSV color space.

separable_conv2d

Separable 2D convolution: apply horizontal then vertical 1D kernels.

sobel

Sobel edge detection: returns (gradient_x, gradient_y).