Struct ComputeKernel 

pub struct ComputeKernel { /* private fields */ }

Collection of CPU compute kernels for image processing.

Every method is a pure function — &self is used only to read KernelConfig; no mutable state is kept.

Implementations

impl ComputeKernel

pub fn new(config: KernelConfig) -> Self

Create a new ComputeKernel with the given configuration.

pub fn default_config() -> Self

Create a new ComputeKernel with default configuration.

pub fn config(&self) -> &KernelConfig

Return a reference to the current configuration.

pub fn rgba_to_yuv420( &self, rgba: &[u8], width: u32, height: u32, ) -> Option<Vec<u8>>

Convert packed RGBA to planar YUV 420.

Layout of the returned buffer:

• Y plane : width * height bytes
• Cb plane : (width/2) * (height/2) bytes
• Cr plane : (width/2) * (height/2) bytes

rgba must have length width * height * 4. Returns None if the input length is unexpected.

pub fn yuv420_to_rgba( &self, yuv: &[u8], width: u32, height: u32, ) -> Option<Vec<u8>>

Convert planar YUV 420 to packed RGBA.

Expects the same memory layout as produced by rgba_to_yuv420. Returns None if the input length is unexpected.

pub fn gaussian_blur( &self, pixels: &[f32], width: u32, height: u32, sigma: f32, ) -> Option<Vec<f32>>

Apply a separable Gaussian blur.

Kernel radius is ceil(3 * sigma). Each pixel channel is treated as an independent f32 sample (grayscale or multi-channel flattened). Returns None if the input length doesn’t match width * height.

pub fn sobel_edges( &self, gray: &[f32], width: u32, height: u32, ) -> Option<Vec<f32>>

Compute Sobel gradient magnitude for a grayscale image.

Input gray must have length width * height. Returns None if length mismatch. Border pixels are set to 0.

pub fn histogram_equalization( &self, gray: &[u8], width: u32, height: u32, ) -> Option<Vec<u8>>

Apply histogram equalization to an 8-bit grayscale image.

Returns None if gray.len() != width * height.

pub fn threshold_otsu( &self, gray: &[u8], width: u32, height: u32, ) -> Option<(u8, Vec<u8>)>

Compute Otsu’s optimal threshold and produce a binary image.

Returns (threshold, binary_image) or None on size mismatch. Binary output: 0 for below-threshold, 255 for at/above.

pub fn alpha_composite( &self, fg: &[u8], bg: &[u8], width: u32, height: u32, ) -> Option<Vec<u8>>

Composite fg over bg using the Porter-Duff “over” operator.

Both buffers must be RGBA, length width * height * 4. Returns None on size mismatch.

pub fn scale_image( &self, pixels: &[u8], src_w: u32, src_h: u32, dst_w: u32, dst_h: u32, ) -> Option<Vec<u8>>

Scale an RGBA image using bilinear interpolation.

Input pixels must be packed RGBA with length src_w * src_h * 4. Returns None on size mismatch or zero dimensions.