Crate yscv_imgproc 

Image processing primitives for yscv.

Structs

BBox

Axis-aligned bounding box. Bounding box with confidence for NMS.

BriefDescriptor

256-bit BRIEF binary descriptor. BRIEF descriptor: 256-bit binary descriptor stored as 32 bytes.

CameraIntrinsics

Camera intrinsic parameters (fx, fy, cx, cy, distortion coefficients). Camera intrinsic parameters including radial and tangential distortion coefficients.

CannyScratch

Pre-allocated scratch buffers for canny_with_scratch. Canny edge detection on a single-channel HWC image.

ComponentStats

Statistics for a connected component region. Statistics for a single connected component.

Contour

Contour represented as a sequence of (row, col) points. A contour as an ordered list of (x, y) boundary pixel coordinates.

DrawDetection

Detection result for drawing (bounding box, label, confidence). A detection result for drawing purposes.

FarnebackConfig

Configuration for Farneback dense optical flow. Configuration for Farneback dense optical flow.

HarrisKeypoint

Harris corner keypoint with location and response strength. Detected corner/interest point.

HoughLine

Hough line in (rho, theta) form. Detected line in Hough parameter space.

ImageF32

HWC f32 image wrapper: (width, height, channels, Vec<f32>). Simple f32 image wrapper (HWC layout, row-major). Zero overhead.

ImageU8

HWC u8 image wrapper: (width, height, channels, Vec<u8>). Simple u8 image wrapper (HWC layout, row-major).

Keypoint

Detected keypoint with location, response, and scale. A detected keypoint with orientation and scale information.

OrbConfig

Configuration for ORB feature detector. Configuration for the ORB feature detector.

OrbDescriptor

ORB binary descriptor (256-bit). ORB descriptor: 256-bit binary descriptor stored as 32 bytes.

OrbFeature

Combined ORB keypoint + descriptor. ORB feature: oriented FAST keypoint paired with a rotated BRIEF descriptor.

RegionProp

Properties of a labeled image region. Properties for a labelled region.

StereoConfig

Configuration for stereo block matching. Configuration for stereo block matching.

SurfKeypoint

SURF keypoint. SURF keypoint with position, scale, orientation, and response.

TemplateMatchResult

Result of template matching (location + score). Template matching result: top-left location and score.

Enums

ImgProcError

TemplateMatchMethod

Method selection for template matching. Template matching method.

Constants

CRATE_ID

Functions

adaptive_threshold_gaussian

Adaptive threshold using Gaussian-weighted local mean. Input: [H, W] grayscale. Adaptive threshold using a Gaussian-weighted local window.

adaptive_threshold_mean

Adaptive threshold using box-filter local mean. Input: [H, W] grayscale. Adaptive threshold using a local mean window.

add_weighted

Alpha-blend two images with weights. Input: [H, W, C] f32. Alpha-blend two images: dst = alpha * a + beta * b + gamma.

adjust_gamma

Gamma correction. Input: [H, W] or [H, W, C]. Applies gamma correction to each pixel: pixel^gamma.

adjust_log

Logarithmic intensity adjustment. Applies logarithmic transform: log(1 + pixel), normalized to [0, 1].

approx_poly_dp

Approximate a contour with fewer vertices (Ramer-Douglas-Peucker). Douglas-Peucker contour approximation.

arc_length

Compute the arc length (perimeter) of a contour. Computes the perimeter (arc length) of a contour.

bilateral_filter

Edge-preserving bilateral filter. Input: [H, W] or [H, W, C]. Bilateral filter on a single-channel [H, W, 1] image.

bilateral_filter_u8

Bilateral filter on u8 image. Bilateral filter operating directly on u8 pixel data.

bitwise_and

Bitwise AND of two binary/grayscale images. Elementwise AND of two single-channel binary masks.

bitwise_not

Bitwise NOT of a binary/grayscale image. Elementwise NOT of a single-channel binary mask.

bitwise_or

Bitwise OR of two binary/grayscale images. Elementwise OR of two single-channel binary masks.

blob_log

Laplacian-of-Gaussian blob detection. Input: [H, W] grayscale. Laplacian-of-Gaussian blob detection on a grayscale [H, W, 1] image.

bounding_rect

Compute the bounding rectangle of a contour. Computes the axis-aligned bounding rectangle of a contour.

box_blur_3x3

3x3 box blur (mean filter). Input: [H, W] or [H, W, C] f32. Applies zero-padded 3x3 box blur over each channel.

box_blur_3x3_f32

3x3 box blur on f32 image. 3x3 box blur on single-channel f32 image. Direct [1,1,1]x[1,1,1]/9 kernel. Replicate border.

box_blur_3x3_u8

3x3 box blur on u8 image. 3x3 box blur on single-channel u8 image. Direct single-pass approach using vextq for horizontal shifts (no intermediate buffer). Division by 9: (sum * 3641 + 16384) >> 15 (3641/32768 ≈ 1/9.005). Edge pixels use replicate (clamp) border.

build_integral_image

Build integral image (summed area table) for SURF. Build integral image for fast box filter computation.

canny

Canny edge detector. Input: [H, W] grayscale f32, thresholds in image intensity range. Returns binary edge map [H, W] with edges = 1.0.

canny_u8

Canny edge detection on u8 grayscale image. Canny edge detection on single-channel u8 image. Full pipeline: Gaussian blur → Sobel L1 magnitude + direction → NMS + threshold → iterative hysteresis. low_thresh and high_thresh are in [0, 255].

canny_with_scratch

Canny edge detection with pre-allocated scratch buffers for zero-alloc hot paths. Canny edge detection with reusable scratch buffers.

center_crop

Center crop to target size. Input: [H, W, C] or [H, W]. Center-crop an [H, W, C] image to size × size.

chw_to_hwc

Convert CHW tensor layout to HWC. Convert an image from [C, H, W] (CHW) layout to [H, W, C] (HWC) layout.

clahe

Contrast-limited adaptive histogram equalization. Input: [H, W] grayscale. Applies CLAHE to a grayscale [H, W, 1] image.

clahe_u8

CLAHE on u8 grayscale image. CLAHE operating entirely in u8 domain — no float conversion in the hot loop.

closing_3x3

Morphological closing (dilate then erode) with 3x3 structuring element. Input: [H, W] grayscale f32. Applies a 3x3 closing (dilate followed by erode) per channel.

color_jitter

Random color jitter augmentation (brightness, contrast, saturation, hue). Apply random color jitter: brightness, contrast, saturation, hue adjustments.

compute_brief

Compute BRIEF binary descriptors at given keypoints. Input: [H, W] grayscale. Compute BRIEF descriptors for keypoints on a grayscale [H, W, 1] image.

compute_gradient_orientation

Compute gradient orientation from Sobel gradients. Computes image gradients and orientation for descriptor construction. Returns a gradient magnitude + orientation pair per pixel.

compute_surf_descriptors

Compute SURF descriptors at detected keypoints. Compute SURF descriptors for detected keypoints.

connected_components_4

4-connected component labeling. Input: [H, W] binary. Connected-component labeling on a binary single-channel HWC image (4-connectivity).

connected_components_with_stats

Connected components with statistics (area, bounding box, centroid). Connected-component labelling with per-component statistics.

contour_area

Compute contour area (signed, using shoelace formula). Computes the area of a polygon defined by its vertices using the Shoelace formula.

convex_hull

Compute the convex hull of a set of 2D points. Computes the convex hull of 2D points using Andrew’s monotone chain algorithm.

corner_sub_pix

Sub-pixel corner refinement. Refines corner locations to sub-pixel accuracy using gradient-based method.

crop

Crop a region from an image tensor. Input: [H, W, C] or [H, W]. Crops an HWC tensor to [crop_h, crop_w, C] starting at (top, left).

dense_optical_flow

Dense optical flow (Farneback method). Input: two consecutive [H, W] grayscale frames. Dense optical flow estimation between two grayscale [H,W,1] frames.

detect_orb

Detect ORB features (keypoints + descriptors). Input: [H, W] grayscale f32. Returns Vec<OrbFeature>. Detect ORB features: oriented FAST keypoints + rotated BRIEF descriptors.

detect_surf_keypoints

Detect SURF keypoints using fast Hessian response. Detect SURF keypoints using box-filter approximation of Hessian.

dilate

Morphological dilation with arbitrary structuring element. Dilate a single-channel [H, W, 1] image with an arbitrary structuring element.

dilate_3x3

Morphological dilation with 3x3 square structuring element. Input: [H, W] grayscale f32. Applies a 3x3 grayscale/RGB dilation (local maximum per channel).

dilate_3x3_f32

3x3 dilation on f32 image. 3x3 dilation (max) on single-channel f32 image.

dilate_3x3_u8

3x3 dilation on u8 image. 3x3 dilation on single-channel u8 image.

distance_transform

Euclidean distance transform. Input: [H, W] binary f32. L1 distance transform on a binary single-channel [H, W, 1] image.

distance_transform_u8

Distance transform on u8 binary image. L1 distance transform on binary u8 image. Non-zero pixels get distance 0, zero pixels get the L1 distance to the nearest non-zero pixel. Returns result as a flat Vec<u16> (max distance = h + w).

draw_circle

Draw a circle on a u8 image. Draw a circle using the midpoint circle algorithm on a [H, W, 3] image tensor.

draw_detections

Draw detection boxes with labels on a u8 image. Draw detections (bounding boxes + labels) on a [H, W, 3] image tensor.

draw_line

Draw a line on a u8 image. Draw a line using Bresenham’s algorithm on a [H, W, 3] image tensor.

draw_polylines

Draw polylines on a u8 image. Draw connected line segments on a [H, W, 3] image tensor.

draw_rect

Draw a rectangle on a u8 image. Draw a rectangle on a [H, W, 3] image tensor (values [0, 1]).

draw_text

Draw text on a u8 image. Draw text on a [H, W, 3] image tensor using an embedded 8×8 bitmap font.

draw_text_scaled

Draw scaled text on a u8 image. Draw text with integer scaling (nearest-neighbor scaled 8x8 bitmap font).

elastic_transform

Elastic deformation augmentation. Elastic distortion transform.

erode

Morphological erosion with arbitrary structuring element. Erode a single-channel [H, W, 1] image with an arbitrary structuring element.

erode_3x3

Morphological erosion with 3x3 square structuring element. Input: [H, W] grayscale f32. Applies a 3x3 grayscale/RGB erosion (local minimum per channel).

erode_3x3_u8

3x3 erosion on u8 image. 3x3 erosion on single-channel u8 image.

farneback_flow

Farneback dense optical flow between two frames. Compute dense optical flow using Farneback’s polynomial expansion method.

fast9_detect

FAST-9 corner detector. Input: [H, W] f32 grayscale. Returns Vec<Keypoint> of detected corners above the threshold. FAST-9 corner detection on a single-channel [H, W, 1] image.

fast9_detect_u8

FAST-9 corner detection on u8 grayscale image. FAST-9 corner detection on u8 single-channel image. Returns keypoints as (x, y) coordinates.

fast_corners

FAST corner detection (configurable arc length). FAST-9 corner detector on a single-channel [H, W, 1] image.

fill_poly

Fill a polygon region on a u8 image. Fill a polygon using scanline fill on a [H, W, 3] image tensor.

filter2d

Generic 2D convolution with a custom kernel. 2D convolution of a single-channel [H, W, 1] image with an arbitrary kernel [KH, KW, 1].

find_contours

Find contours in a binary image. Input: [H, W] binary. Finds external contours in a binary single-channel [H, W, 1] image.

fit_ellipse

Fit an ellipse to a set of 2D points (least-squares). Fits an axis-aligned ellipse to a set of 2D points using the method of moments.

flip_horizontal

Flip image horizontally. Input: [H, W, C] or [H, W]. Flips an HWC tensor horizontally (mirror over vertical axis).

flip_vertical

Flip image vertically. Input: [H, W, C] or [H, W]. Flips an HWC tensor vertically (mirror over horizontal axis).

gaussian_blur_3x3

3x3 Gaussian blur (sigma approx 0.85). Input: [H, W] or [H, W, C] f32. Applies zero-padded 3x3 Gaussian blur per channel. Kernel: [[1,2,1],[2,4,2],[1,2,1]] / 16.

gaussian_blur_3x3_f32

3x3 Gaussian blur on f32 image. 3x3 Gaussian blur on single-channel f32 image. Direct [1,2,1]x[1,2,1]/16 kernel. Replicate border.

gaussian_blur_3x3_u8

3x3 Gaussian blur on u8 image. 3x3 Gaussian blur on single-channel u8 image. Uses [1,2,1] separable kernel with u16 intermediate. Tiled processing: H-pass and V-pass are fused within tiles of 8 rows, so the u16 tile buffer (10 rows * w * 2 bytes) fits in L1 cache. Rayon parallelism across tiles when image is large enough.

gaussian_blur_5x5

5x5 Gaussian blur (sigma approx 1.0). Input: [H, W] or [H, W, C] f32. Applies zero-padded 5x5 Gaussian blur per channel.

gaussian_pyramid

Gaussian image pyramid (successive half-resolution blurred images). Builds a Gaussian image pyramid by repeated 2x downsampling.

good_features_to_track

Shi-Tomasi / Good Features to Track corner detector. Shi-Tomasi corner detection (minimum eigenvalue approach).

grayscale_f32

Convert f32 RGB image to grayscale. Converts RGB f32 image [H,W,3] to grayscale [H,W,1]. Uses BT.601: gray = 0.299R + 0.587G + 0.114*B.

grayscale_u8

Convert u8 RGB image to grayscale.

hamming_distance

Hamming distance between two binary descriptors. Hamming distance between two BRIEF descriptors.

harris_corners

Harris corner detector. Input: [H, W] grayscale f32. Returns corner response map [H, W]. Harris corner detector on a single-channel [H, W, 1] image.

harris_corners_u8

Harris corner detection on u8 image. Harris corner detection operating directly on u8 grayscale image data.

histogram_256

Compute 256-bin grayscale histogram. Input: [H, W] f32 with values in [0, 255]. Computes a 256-bin histogram for a single-channel [H, W, 1] image. Assumes values in [0, 1] range.

histogram_equalize

Histogram equalization. Input: [H, W] grayscale. Histogram equalization for single-channel [H, W, 1] images with values in [0, 1].

histogram_u8

Compute 256-bin histogram from u8 grayscale data. Computes a 256-bin histogram from raw u8 data.

hog_cell_descriptor

HOG (Histogram of Oriented Gradients) cell descriptor. Histogram of Oriented Gradients (HOG) descriptor for a single cell.

homography_4pt

Compute 4-point homography (direct linear transform). Computes a 3x3 homography matrix from 4 source/destination point correspondences using the Direct Linear Transform (DLT) algorithm.

hough_circles

Hough circle detection. Input: [H, W] edge/binary image. Hough circle transform on a binary/edge single-channel [H, W, 1] image.

hough_lines

Hough line detection. Input: [H, W] edge/binary image. Standard Hough line transform on a binary/edge single-channel [H, W, 1] image.

hsv_to_rgb

Convert HSV to RGB. Input/output: [H, W, 3] f32. Converts HSV [H, W, 3] to RGB [H, W, 3]. Expects H in [0, 1], S in [0, 1], V in [0, 1].

hu_moments

Compute Hu’s seven moment invariants from a binary/grayscale image. Computes the 7 Hu invariant moments from a single-channel [H, W, 1] image.

hwc_to_chw

Convert HWC tensor layout to CHW. Convert an image from [H, W, C] (HWC) layout to [C, H, W] (CHW) layout.

imagenet_preprocess

ImageNet-standard preprocessing (resize 256, center crop 224, normalize). Standard ImageNet preprocessing pipeline.

imread

Read an image file as ImageU8. Supports PNG, JPEG, BMP, GIF, TIFF, WebP. Read an image from disk and return it as a [H, W, C] f32 tensor with values in [0, 1].

imread_gray

Read an image file as grayscale ImageU8. Read an image as grayscale and return it as a [H, W, 1] f32 tensor with values in [0, 1].

imwrite

Write an ImageU8 to a file. Format inferred from extension (png, jpg, bmp, etc.). Write a [H, W, C] f32 tensor (values in [0, 1]) to disk.

in_range

Check if pixel values fall within a per-channel range. Create a binary mask where each pixel is 1.0 if all channels are within [lower, upper] (inclusive), 0.0 otherwise.

inpaint_telea

Inpaint missing regions using Telea’s fast marching method. Telea-style fast marching inpainting.

integral_image

Compute integral image (summed area table). Input: [H, W]. Computes the integral (summed-area table) image from a [H, W, 1] tensor.

lab_to_rgb

Convert CIE Lab* to RGB. Input/output: [H, W, 3]. Converts CIE LAB [H, W, 3] to RGB [H, W, 3].

laplacian_3x3

3x3 Laplacian filter. Input: [H, W] grayscale f32. Applies 3x3 Laplacian edge detection per channel.

lucas_kanade_optical_flow

Lucas-Kanade sparse optical flow. Sparse Lucas-Kanade optical flow between two grayscale [H,W,1] frames.

match_features

Match feature descriptors between two sets by brute-force nearest neighbor. Match ORB features between two sets using brute-force Hamming distance.

match_surf_descriptors

Match two sets of SURF descriptors. Match SURF descriptors between two sets using nearest-neighbor ratio test.

median_blur_3x3

3x3 median blur. Input: [H, W] or [H, W, C] f32. Applies 3x3 median filter per channel.

median_blur_3x3_u8

3x3 median blur on u8 image. 3x3 median filter on single-channel u8 image. 3x3 median blur on single-channel u8 image. Uses a SIMD sorting network (NEON) to find the median of 9 elements, processing 16 pixels at a time. Scalar fallback for non-NEON and border pixels.

median_filter

Median filter with arbitrary window size. Applies NxN median filter on a single-channel [H, W, 1] image.

min_area_rect

Minimum area rotated rectangle of a set of 2D points. Computes the minimum-area bounding rectangle for a set of 2D points using the rotating calipers approach on the convex hull.

morph_blackhat

Morphological black-hat transform (closing - input). Black-hat transform: closing(input) - input.

morph_gradient_3x3

Morphological gradient (dilation - erosion) with 3x3 element. Applies a 3x3 morphological gradient (dilate - erode) per channel.

morph_tophat

Morphological top-hat transform (input - opening). Top-hat transform: input - opening(input).

nms

Non-maximum suppression on bounding boxes by IoU threshold. Greedy non-maximum suppression on bounding boxes.

normalize

Normalize image channels: (pixel - mean) / std. Input: [H, W, C] f32. mean and std slices must have length equal to C. Per-channel normalize in HWC layout: (x - mean[c]) / std[c].

normalize_image

Normalize u8 image to [0, 1] f32 then apply mean/std normalization. Per-channel normalization: (pixel - mean[c]) / std[c] for an HWC image.

opening_3x3

Morphological opening (erode then dilate) with 3x3 structuring element. Input: [H, W] grayscale f32. Applies a 3x3 opening (erode followed by dilate) per channel.

orb_descriptors

Compute ORB descriptors at given keypoints. Computes ORB descriptors for keypoints on a grayscale [H,W,1] image.

orb_hamming_distance

Hamming distance between two ORB descriptors. Hamming distance between two ORB descriptors.

orb_match

Match ORB features between two sets. Brute-force ORB descriptor matching. Returns (idx_a, idx_b, distance).

pad_constant

Pad image borders with a constant value. Pads an HWC tensor with a constant value on all sides.

project_points

Project 3D points to 2D using camera intrinsics. Project 3-D world points to 2-D pixel coordinates using a pinhole camera model (no extrinsic rotation/translation — assumes camera-frame coordinates).

random_crop

Random crop augmentation. Randomly crop a [H, W, C] image to (out_h, out_w).

random_erasing

Random erasing (cutout) augmentation. Random erasing (cutout): randomly erase a rectangular region with probability p.

random_horizontal_flip

Random horizontal flip augmentation (50% probability). Randomly flip horizontally with probability p.

random_rotation

Random rotation augmentation. Rotate image by a random angle in [-max_degrees, max_degrees].

random_vertical_flip

Random vertical flip augmentation (50% probability). Randomly flip vertically with probability p.

ransac_homography

Estimate homography via RANSAC. RANSAC-based homography estimation from point correspondences.

region_props

Compute region properties (area, centroid, bbox) from a labeled image. Compute region properties from a label image.

remove_small_objects

Remove small connected components below a size threshold. Removes connected components with area less than min_size.

rescale_intensity

Rescale intensity to [0, 1] or a target range. Linearly rescales pixel intensities from [in_min, in_max] to [out_min, out_max].

resize_bilinear

Resize using bilinear interpolation. Input: [H, W, C] or [H, W] f32. Resizes HWC tensor with bilinear interpolation.

resize_bilinear_u8

Bilinear resize on u8 image. Bilinear resize on u8 image [H,W,C] to target dimensions. Uses fixed-point arithmetic (8-bit fractional) for speed. Single-channel: NEON gather + blend, 8 pixels at a time. Multi-channel: separable H-pass (scalar gather into u16 buf) then NEON-vectorized V-pass on contiguous data, with row caching.

resize_nearest

Resize using nearest-neighbor interpolation. Input: [H, W, C] or [H, W] f32. Resizes HWC tensor with nearest-neighbor sampling.

resize_nearest_u8

Nearest-neighbor resize on u8 image. Nearest-neighbour resize on u8 image [H,W,C] to target dimensions.

rgb_to_bgr

Convert RGB to BGR channel order. Input/output: [H, W, 3]. Converts RGB [H, W, 3] to BGR [H, W, 3] (channel swap).

rgb_to_grayscale

Convert RGB to grayscale using ITU-R BT.601 luminance weights. Input: [H, W, 3], output: [H, W]. Converts RGB image [H, W, 3] to grayscale [H, W, 1].

rgb_to_hsv

Convert RGB to HSV color space. Input/output: [H, W, 3] f32. Converts RGB [H, W, 3] to HSV [H, W, 3]. H is in [0, 1] (scaled from [0, 360]), S and V are in [0, 1].

rgb_to_hsv_u8

RGB to HSV conversion on u8 image. Converts RGB u8 image to HSV u8 image.

rgb_to_lab

Convert RGB to CIE Lab*. Input/output: [H, W, 3]. Converts RGB [H, W, 3] to CIE LAB [H, W, 3].

rgb_to_yuv

Convert RGB to YUV. Input/output: [H, W, 3]. Converts RGB [H, W, 3] to YUV [H, W, 3] using BT.601 coefficients.

rotate90_cw

Rotate image 90 degrees clockwise. Rotates an HWC tensor 90 degrees clockwise.

scharr_3x3_gradients

3x3 Scharr gradient (dx, dy). Input: [H, W] grayscale. Computes 3x3 Scharr gradients (gx, gy) per channel.

scharr_3x3_magnitude

3x3 Scharr gradient magnitude. Input: [H, W] grayscale. Computes the Scharr gradient magnitude: sqrt(gx^2 + gy^2).

sift_descriptor

Compute SIFT-style descriptor at a keypoint. SIFT-like 128-element descriptor for a keypoint on a grayscale [H,W,1] image.

sift_match

Match SIFT descriptors. Matches SIFT descriptors using Euclidean distance with Lowe’s ratio test.

skeletonize

Morphological skeletonization. Input: [H, W] binary. Zhang-Suen thinning algorithm on a binary single-channel [H, W, 1] image.

sobel_3x3_f32

3x3 Sobel gradient magnitude on f32 image. 3x3 Sobel gradient magnitude on single-channel f32 image. Returns |Gx| + |Gy| approximation. Border pixels are zero.

sobel_3x3_gradients

3x3 Sobel gradient (dx, dy). Input: [H, W] grayscale f32. Computes 3x3 Sobel gradients (gx, gy) per channel.

sobel_3x3_magnitude

3x3 Sobel gradient magnitude. Input: [H, W] grayscale f32. Computes 3x3 Sobel gradient magnitude sqrt(gx^2 + gy^2) per channel.

sobel_3x3_magnitude_u8

3x3 Sobel gradient magnitude on u8 image. 3x3 Sobel gradient magnitude on single-channel u8 image. Returns gradient magnitude clamped to [0, 255].

stereo_block_matching

Stereo block matching for disparity estimation. Compute a disparity map from a rectified stereo pair using block matching (SAD).

template_match

Template matching. Input: [H, W] image and [TH, TW] template. Slides template over image computing a similarity map, returns the best match.

threshold_binary

Binary threshold: output 1.0 where input >= thresh, else 0.0. Input: [H, W]. Binary threshold: outputs max_val where value > threshold, else 0.0.

threshold_binary_f32

Binary threshold on f32 image. Binary threshold on single-channel f32 image. Pixels > thresh become max_val, otherwise 0.

threshold_binary_inv

Inverse binary threshold: output 1.0 where input < thresh. Input: [H, W]. Inverse binary threshold: outputs 0.0 where value > threshold, else max_val.

threshold_otsu

Otsu’s automatic threshold. Input: [H, W] grayscale with values in [0, 255]. Otsu threshold for single-channel [H, W, 1] images. Returns (threshold, thresholded_image).

threshold_truncate

Truncate threshold: clamp values above thresh. Input: [H, W]. Truncate threshold: caps values above threshold.

triangulate_points

Triangulate 3D points from two camera views. Triangulate 3-D points from corresponding 2-D observations in a rectified stereo pair.

undistort_points

Undistort 2D points using camera intrinsics. Remove lens distortion from a set of 2-D pixel coordinates.

warp_affine

Apply an affine warp transform. Input: [H, W, C] or [H, W]. Applies a 2x3 affine transformation to an HWC image using bilinear interpolation.

warp_perspective

Apply a perspective warp (homography). Input: [H, W, C] or [H, W]. Applies a 3x3 perspective (homography) transformation to an [H, W, C] image.

warp_perspective_u8

Perspective warp on u8 image. Warp u8 single-channel image using a 3x3 homography matrix. Uses bilinear interpolation. Out-of-bounds pixels are set to 0.

watershed

Watershed segmentation. Input: [H, W] grayscale, markers: [H, W] integer labels. Marker-based watershed segmentation on a grayscale [H,W,1] image.

yuv_to_rgb

Convert YUV to RGB. Input/output: [H, W, 3]. Converts YUV [H, W, 3] to RGB [H, W, 3] using BT.601 coefficients.

Type Aliases

SurfDescriptor

SURF descriptor. SURF descriptor (64-element vector).