Image Filters & Blur

Image filtering is essential for noise reduction, smoothing, and feature enhancement. Iris implements standard spatial filters using parallelized row-by-row CPU iterations (powered by rayon under the parallel feature flag) or accelerated GPU tensor calculations.

Available Blur Filters

Box Blur

Smooths an image using a normalized box filter of the given kernel size.

// Applies a 5x5 box filter
let blurred = image.box_blur(5)?;

Gaussian Blur

Smooths an image using a Gaussian kernel. You must specify the kernel size (must be odd) and standard deviation sigma.

// Applies a 5x5 Gaussian blur with sigma = 1.5
let gaussian = image.gaussian_blur(5, 1.5)?;

Median Blur

Reduces salt-and-pepper noise by taking the median value in each local neighborhood.

// Applies a 5x5 median filter
let median = image.median_blur(5)?;

Bilateral Filter

Smooths the image while preserving sharp edges. It uses a range sigma for color similarity and space sigma for coordinate closeness.

// Applies a bilateral filter with d=5, sigma_color=0.1, sigma_space=10.0
let filtered = image.bilateral_filter(5, 0.1, 10.0)?;

Separable 2D Filter

Applies two 1D kernels sequentially along the X and Y dimensions to achieve efficient custom 2D filtering.

let kernel_x = vec![0.25, 0.5, 0.25];
let kernel_y = vec![0.25, 0.5, 0.25];
let filtered = image.sep_filter_2d(&kernel_x, &kernel_y)?;

Utility Filters

Filter2D

Applies a general 2D convolution kernel to the image.

let kernel: &[&[f32]] = &[
    &[0.0, -1.0, 0.0],
    &[-1.0, 5.0, -1.0],
    &[0.0, -1.0, 0.0],
];
let sharpened = image.filter2d(kernel, None, 0.0)?;

Add Weighted

Blends two images: result = alpha * src1 + beta * src2 + gamma.

let blended = img1.add_weighted(&img2, 0.7, 0.3, 0.0)?;

Convert Scale Abs

Converts the image with per-pixel scale and shift, then takes absolute values.

let abs_img = image.convert_scale_abs(1.0, 0.0)?;

Distance Transform

Computes the distance transform of a binary image, where each pixel value is its distance to the nearest zero pixel.

let dist = image.distance_transform()?;