IPCV

IPCV is an Image Processing and Computer Vision toolbox for Scilab. It provides Scilab macros and native gateways for image I/O, filtering, morphology, feature detection, DNN inference, video/camera handling, object tracking, and other OpenCV-backed workflows.

The current development line is IPCV 5.0.0, based on OpenCV/OpenCV contrib 5.0.0 and Scilab 2026.1. The native implementation uses category-based C++ source modules and a current Scilab gateway API.

Rapid development is organized as numbered roadmap steps on master. Short-lived branches may be used for substantial or experimental changes, then merged and deleted. Current evidence, remaining gates, and longer-term priorities are maintained in ROADMAP.md.

Highlights

OpenCV 5 native backend with category-based C++ source files under src/cpp.
Scilab gateway image exchange rewritten for the current Scilab API.
DNN support updated for OpenCV 5, including ONNX examples and OpenCV Zoo model download support.
dnn_forward_preset provides reusable ImageNet, CLIP, and OpenCV Zoo preprocessing configurations, including per-channel standard deviation.
dnn_forward_outputs runs named or unconnected output layers and returns tensors in a list with matching layer names.
OpenCV Zoo browser GUI: opencv_zoo_gui().
Step 3 thresholding, binarization, connected-component measurement, binary geometry, reconstruction, extrema, Sobel gradients, filtering, denoising, and local-statistics APIs.
Foundation help examples now show threshold, binary-mask, connected-component, label, area-filter, and border-cleanup results on OpenCV sample images.
Batch 5 help examples now use OpenCV sample images for filtering, enhancement, gradients, extrema, reconstruction, and binary measurement workflows.
Step 3 image workflows: flipping, padding, tiling, bottom-hat filtering, k-means segmentation, GrabCut, superpixels, region properties, entropy, range, variance, order-statistic filtering, callback neighborhood filtering, and flat-field correction.
Batch 6 help examples now use OpenCV sample images and show source/result comparisons for the main workflows.
Step 3 Batch 7 adds translation, pairwise display comparison, contrast limits, local brightening, grayscale region growing, multilevel quantization, binary topology and geodesic distance, inverse Radon, Hough peak/line conversion, corners, HOG, and regularized deconvolution helpers.
Batch 7 help examples now use OpenCV sample images and show source/result or score-map comparisons for the main workflows.
Step 3 Batch 8 adds histogram matching, local contrast, extrema suppression, marker minima, ROI masks, texture maps, mean-shift-like smoothing, template matching, histogram back-projection, optical-flow translation fields, lens correction, and projective rectification.
Batch 8 help examples now use the OpenCV smarties, fruits, and chessboard sample assets with before/after or score-map output where appropriate.
Step 3 Batch 9 adds flood fill, local range, color transfer, diffusion, guided filtering, local normalization, seamless compositing, phase-correlation registration, camera-calibration initialization, stereo BM/SGBM baselines, and line/ellipse fitting. Each new public function includes a runnable help example, author metadata, and History.
Batch 9 examples use the OpenCV chessboard, smarties, and fruits sample assets where appropriate; stereo and calibration examples identify their illustrative input assumptions.
Proposed duplicate names are intentionally not added: use rgb2gray or immat2gray instead of im2gray, imoverlaymask instead of imoverlay, imlabel2rgb with overlay helpers instead of imlabeloverlay, imphasecorr instead of imregcorr, and imtransform instead of imwarp.
Step 3 Batch 10 adds robust statistics, FCM and active-contour segmentation, phase-congruency and ridge maps, 3D volume operations, and OpenCV 5 camera geometry with imsolvepnp, imestimatefundamental, and imtriangulate. Circle detection remains available through the existing imhoughc API.
Step 3 Batch 11 adds adaptive thresholds, gradient weights, GLCM texture statistics, LBP descriptors, image moments, orientation and Feret measurements, local variance, color masks, aspect-ratio crop, and 3D intensity/filtering helpers.
Step 3 Batch 12 adds im2uint32, imlut, imapplycolormap, imintegral, imbwdist, imgraydist, imbwulterode, imlocallapfilt, imreducehaze, improfile3, imtranslate3, imregionalmax3, imbwmorph3, imminarearect, and imminenclosingcircle.
Batch 12 examples use the OpenCV fruits and smarties images, synthetic volumes, and plotted point sets so each new function has a visible workflow.
Batch 12 intentionally skips duplicate or near-duplicate proposals: im2gray because rgb2gray already covers the conversion, imstdfilt because imlocalstd exists, imentropyfilt because imlocalentropy exists, imrangefilt because imlocalrange exists, imwarp because imtransform and the affine/perspective helpers exist, and imregcorr because imphasecorr exists.
Step 3 Batch 13 adds 2-D/3-D binary component and boundary helpers, image-reference and registration configuration structures, 3-D k-means segmentation, bitwise and remapping utilities, and Harris, KAZE, and AKAZE feature workflows.
Batch 13 examples follow OpenCV workflows for Harris corner detection, AKAZE/KAZE local features, contour analysis, and connected-component filtering. im2gray, imstdfilt, imentropyfilt, imrangefilt, imnormxcorr2, imwarp, imregcorr, and imbwskel remain intentionally skipped because active IPCV equivalents already exist.
Batch 13 help examples now show their source reference and provide two independent executable blocks: an adapted OpenCV workflow and a separate IPCV workflow.
Browser-based imtool() supports images and OpenCV-decoded video, including paused-frame inspection, zoom, pan, rotation, color-space inspection, pixel grids, temporal difference/heatmap/average analysis, optical-flow overlays, and recording of the rendered analysis stream.
Native OpenCV 5 gateways now cover k-means segmentation, GrabCut, superpixels, Harris/KAZE/AKAZE features, KAZE/AKAZE descriptors, template matching, phase registration, solve-PnP, fundamental-matrix estimation, and triangulation.
Volumetric functions are grouped under the dedicated 3D Image Processing category, including filtering, morphology, profiles, transforms, region measurements, segmentation, imshow3d, and the interactive WebGL volshow renderer.
The native 3-D foundation provides unambiguous Scilab hypermatrix exchange, 6/18/26-connected labeling and morphology, plateau-aware regional maxima, and accelerated box, Gaussian, and median filtering.
dicomread() imports uncompressed grayscale single- and multi-frame DICOM data with acquisition metadata; dicomshow() displays selected frames with per-slice, shared, or hidden information panels.
volshow() provides GPU ray-cast volume compositing, maximum-intensity projection, shaded isosurfaces, rotatable orthogonal slice planes, scrollable 2-D layers, and full-resolution layer export to the Scilab workspace with DICOM voxel spacing.
Display and geometry workflows consistently distinguish upper-left image coordinates from lower-left Scilab plot coordinates. Help overlays use rect2cart() or sub2cart() where required.
The help audit replaces toy or misleading examples with executable image workflows, adds missing Parameters/Author/History sections, removes stale duplicate pages, and links examples to their actual OpenCV references.
AVI examples use explicit writable output paths and the broadly compatible MJPG codec. addframe remains available as a deprecated compatibility alias of aviaddframe.
Public im... names are canonical in the OpenCV 5 development line; archived MATLAB-style aliases are retained only under macros/old.
Duplicate MATLAB-style help pages and stale examples are removed; help links now target the active im... replacements.
Current OpenCV 5 migration tests are in tests/unit_tests; legacy 4.5 tests are kept only as an optional archive/reference set.
Large DNN .onnx models are excluded from git and downloaded on demand.

Current Development Status

The completed codex/new-features work was merged into master through PR #31, and that feature branch has been retired. On 17-Jul-26, the toolbox completed a clean VS 2022 x64 native build, repository validation, regenerated help, and the Windows stability suite with 35 passed and 0 failed. Release preparation still requires the broader release/GUI/network/hardware suites where applicable and macOS/Linux validation. Generated binaries, help output, downloaded models, and local Codex/Scilab profiles are not source-controlled.

Codex And GPT-5.6 During OpenAI Build Week

IPCV 5.0 development was accelerated with Codex using GPT-5.6. The largest migration task was rebasing IPCV's native backend from OpenCV 4.5 to OpenCV 5.0. Codex helped inspect the legacy Scilab and C++ gateways, trace changed or removed OpenCV APIs, reorganize the native boundary, update the build workflow, and repeatedly compile, diagnose, and validate the migration. The core rebase reached a working state in about one day; based on prior manual maintenance of IPCV, the same investigation and migration would normally take weeks and could extend to months across hundreds of functions and multiple platforms.

Codex and GPT-5.6 accelerated the engineering loop for:

Migrating and refactoring OpenCV-backed gateways and function families.
Expanding Scilab functions, help pages, runnable examples, and regression coverage.
Diagnosing build, API-compatibility, packaging, GUI, video, DNN, DICOM, and three-dimensional volume-processing issues.
Creating the containerized browser demonstration so judges can run curated IPCV operations without installing the full desktop stack.
Preparing validation evidence, project media, and technical documentation.

The active Scilab function-file count grew from 196 in IPCV 4.5 to 477 in the current IPCV 5.0 development tree: a net increase of 281 functions, or 143.4% growth (2.43x). The official Build Week baseline contained 326 active function files; 151 were added during the event period.

This was not a one-click conversion. Human judgment remained responsible for the architecture, public API, compatibility requirements, educational direction, and definition of correct behavior. Codex and GPT-5.6 shortened the inspect-implement-build-test-document cycle while those decisions remained under project-owner control.

Coordinate Convention

Image-processing functions use image coordinates: [x y] means column and row, with the origin at the upper-left image corner. Scilab plot() uses Cartesian coordinates with the origin at the lower-left of the displayed axes. When an example overlays points, lines, or contours on an image shown with imshow(), it must convert image coordinates with rect2cart() or sub2cart() before calling plot().

OpenCV Example Assets

Current help examples use image assets from the official OpenCV sample-data repository:

images/opencv_smarties.png: color segmentation, statistics, and volume examples.
images/dicom/emri_small.dcm: anonymized 10-frame MR sample from the MIT-licensed pydicom-data repository; attribution and license are stored beside the file.
images/opencv_fruits.jpg: color/intensity statistics and gradient examples.
images/opencv_left01.jpg and images/opencv_left02.jpg: chessboard pose, stereo, and camera-geometry examples.

The workflows are adapted from OpenCV's camera calibration tutorial, pose estimation tutorial, and epipolar geometry tutorial.

Requirements

Scilab 2026.x
Windows build tools for native compilation
OpenCV/OpenCV contrib 5.0.0 thirdparty build for the IPCV native gateway

For native builds on Windows, start Scilab from a Visual Studio native tools command prompt so the compiler environment is available.

Naming And Archived Aliases

The public IPCV names now use the im... convention. Legacy non-im names remain in macros/old for reference only and are intentionally excluded from the generated macro libraries.

Archived name	Public IPCV name
`edge`	`imedge`
`filter2`	`imfilter2`
`fspecial`	`imfspecial`
`mean2`, `std2`, `stdev2`, `corr2`	`immean2`, `imstd2`, `imstdev2`, `imcorr2`
`mat2gray`	`immat2gray`
`label2rgb`	`imlabel2rgb`
`poly2mask`	`impoly2mask`
`bwareaopen`, `bwareafilt`	`imbwareaopen`, `imbwareafilt`
`bwperim`, `bwselect`, `bwhitmiss`, `bwthin`, `bwmorph`, `bwboundaries`	`imbwperim`, `imbwselect`, `imbwhitmiss`, `imbwthin`, `imbwmorph`, `imbwboundaries`
`bwlabel`, `bwconncomp`, `bwfill`	`imlabel`, `imconnectedcomponents`, `imfill` or `imfillholes`
`adapthisteq`, `wiener2`, `medfilt2`	`imadapthistequal`, `imwiener2`, `immedian`

Functions such as imsharpen, imlaplacian, imgradientdirection, imlocalmean, and imlocalstd are separate compositions, not aliases.

Versioning

IPCV public versions follow the OpenCV baseline used by the native backend. Rapid development work does not increment the package version after every change; it is tracked as Step 1, Step 2, and so on in the roadmap. VERSION and DESCRIPTION change only when a release candidate is deliberately prepared.

The loaded package and native backend versions can be inspected at runtime:

info = ipcv_version();
disp(info);

Load IPCV

From Scilab:

exec("loader.sce", -1);

Then call IPCV functions directly, for example:

S = imread(fullpath(getIPCVpath() + "/images/baboon.png"));
imshow(S);

Build From Source

Build order:

Prepare compiler tools.
Build OpenCV/OpenCV contrib 5.0.0 into thirdparty.
Start Scilab from the compiler environment.
Run IPCV builder.sce.
Load IPCV and run a quick smoke test.

1. Prepare Compiler Tools

On Windows, install Visual Studio 2022 Build Tools or Visual Studio with the C++ workload. Make sure cmake, ninja, curl, and tar are available in the command prompt used for the build.

On Linux or macOS, install CMake, Ninja, a C/C++ compiler toolchain, curl, make, and common build utilities.

2. Build OpenCV 5 Thirdparty

IPCV expects a local OpenCV/OpenCV contrib 5.0.0 build under thirdparty. The helper scripts in thirdparty/build download OpenCV sources, configure CMake with Ninja, build opencv_world, and install headers/libraries into the platform-specific thirdparty folder.

For a detailed Windows and Linux thirdparty build guide, see thirdparty/build/README.txt.

Windows

Open a Visual Studio x64 native tools command prompt, then run:

cd thirdparty\build
build.bat

The script:

Downloads OpenCV and OpenCV contrib 5.0.0.
Applies the IPCV local MLAS skip patch for the OpenCV DNN Windows build.
Configures OpenCV with -G Ninja.
Builds and installs into thirdparty\Windows\%PROCESSOR_ARCHITECTURE%.
Copies generated OpenCV DLL/import-library files into the local lib folder expected by IPCV.

Expected Windows layout after installation:

thirdparty/Windows/AMD64/include/
thirdparty/Windows/AMD64/lib/

Linux And macOS

Run:

cd thirdparty/build
./build.sh

The script first builds a local shared FFmpeg dependency, then downloads and builds OpenCV/OpenCV contrib 5.0.0.

Expected layout after installation:

thirdparty/Linux/<architecture>/
thirdparty/Darwin/<architecture>/

3. Start Scilab From The Compiler Environment

On Windows, keep using the Visual Studio x64 native tools command prompt and launch Scilab from there, for example:

"<SCILAB_INSTALL_DIR>\bin\WScilex.exe"

Using the native tools prompt matters because IPCV builds C++ gateways and must see the same MSVC compiler environment used by OpenCV.

On Linux or macOS, start Scilab from a shell where the compiler, CMake, and OpenCV thirdparty library paths are available.

4. Build IPCV

In Scilab, change to the IPCV root folder and run:

cd("<IPCV_ROOT>");
exec("builder.sce", -1);

The top-level builder compiles macros, native source code, gateways, help, loader, and cleaner files. The source build is organized around src/cpp/builder_cpp.sce, which builds the native libipcv_core library from category files such as ipcv_image_io.cpp, ipcv_filtering.cpp, ipcv_feature_detection.cpp, and ipcv_dnn.cpp.

Expected generated build outputs include:

src/cpp/libipcv_core.*
sci_gateway/cpp/gw_ipcv.*
loader.sce

5. Load And Smoke Test

After the build completes, load IPCV:

exec("loader.sce", -1);

Then run a small image I/O and display check:

S = imread(fullpath(getIPCVpath() + "/images/baboon.png"));
disp(size(S));
imshow(S);

For a focused command-line check, run a current unit test from tests/unit_tests, for example:

test_run(getIPCVpath(), "ipcv_image_io");

Graphical Tools

IPCV includes two browser-based Scilab GUIs:

Image And Video Tool

Launch imtool with an empty workspace, an image matrix, or a media filename:

imtool();

image = imread(fullpath(getIPCVpath() + "/images/baboon.png"));
imtool(image);

imtool("C:/path/to/video.mp4");

The GUI provides:

Image and video loading through IPCV and OpenCV.
Zoom, pan, rotation, fit-to-window, and pixel-grid inspection.
Per-pixel RGB values when the image is zoomed sufficiently.
RGB, grayscale, HSV, Lab, and YCbCr color-space views.
Video playback, pause, seek, frame stepping, and playback-rate controls.
Paused-frame inspection and temporal difference, motion heatmap, temporal average, and sparse optical-flow analysis.
Recording of the rendered analysis stream at 0.25x, 0.5x, 1x, or 2x output speed.

MP4, H.264, and HEVC playback depends on the codecs enabled in the bundled OpenCV build.

OpenCV Zoo Model Browser

IPCV includes a Scilab browser GUI for OpenCV Zoo models:

opencv_zoo_gui();

The GUI can:

Fetch available ONNX models from the official OpenCV Zoo.
Group models by category.
Show real Git LFS model sizes.
Download a selected model.
Download a selected model, generate a starter Scilab script, and open it in the Scilab editor.

Downloaded .onnx files are ignored by git. Starter scripts generated by the OpenCV Zoo GUI in the default DNN download folder are also ignored, so users can create and edit local model-specific experiments without changing the repository. The default download folder is:

images/dnn

DNN Example

dnn_path = fullpath(getIPCVpath() + "/images/dnn/");
model_name = "image_classification_mobilenetv2_2022apr.onnx";
model_file = dnn_path + model_name;

if ~isfile(model_file) then
    model_url = "https://github.com/opencv/opencv_zoo/raw/main/models/image_classification_mobilenet/" + model_name;
    http_get(model_url, model_file, follow=%t, timeout=300);
end

net = dnn_readmodel(model_file, "", "onnx");
net = dnn_setpreferable(net, "opencv", "cpu");

S = imread(fullpath(getIPCVpath() + "/images/baboon.png"));
out = dnn_forward(net, S, [224, 224], [], 1 / 127.5, [127.5 127.5 127.5], 1, 0);

[score, index] = max(out);
disp([index, score]);
dnn_unloadmodel(net);

Tests

Current OpenCV 5 migration tests:

tests/unit_tests

These are the active tests for the C++/OpenCV 5 code path and are the default place for new regression coverage.

Legacy IPCV 4.5 tests are not part of the default 5.0.0 test run. If they are kept for migration comparison, keep them isolated under:

tests/unit_tests/4.5

That archive path is only for historical behavior checks and should not be mixed with the active OpenCV 5 .tst and .dia.ref files.

Run tests from Scilab after loading/building the toolbox as needed. For focused migration checks, run the relevant .tst file directly or use Scilab's test_run workflow.

Tests are classified as core, integration, gui, network, or hardware. The stability suite is the focused migration signal and release combines the deterministic core and local-asset integration tests. Network, GUI, and physical-camera tests are opt-in.

On Windows, run a suite with:

powershell -ExecutionPolicy Bypass -File tests/run_tests.ps1 -ScilabExecutable "C:\Tools\scilab-2026.1.0\bin\WScilex-cli.exe" -Suite stability

On Linux or macOS, run:

sh ./tests/run_tests.sh /path/to/scilab-cli stability

Available suite names are core, integration, gui, network, hardware, stability, release, and all. Reports are written to tests/results and are ignored by git. Run one or more named tests by setting IPCV_TEST_NAMES, or on Windows by passing -TestNames "ipcv_image_io,ipcv_codec_matrix".

The original run_stability_tests.ps1 and run_stability_tests.sce files are kept as compatibility entry points and delegate to the suite runner.

Run the Windows repository validator and deterministic release suite together:

powershell -ExecutionPolicy Bypass -File tests/run_release_checks.ps1 -ScilabExecutable "C:\Tools\scilab-2026.1.0\bin\WScilex-cli.exe"

Add -Build when the command is launched from a Visual Studio x64 native tools environment and a complete native rebuild is required. See tests/README.md for suite definitions, opt-in requirements, timeouts, reports, and test-authoring rules.

Repository Layout

macros/              Scilab user-facing functions grouped by help category
macros/GUI/          Browser/UI assets and GUI-specific macros
sci_gateway/cpp/     Scilab gateway wrappers
src/cpp/             OpenCV-backed C++ implementation
help/en_US/          Help pages
demos/               Demo scripts
tests/unit_tests/    Current unit tests
thirdparty/          Local thirdparty build area
images/              Sample images and downloaded model location

Change History

Detailed release notes, migration notes, and the converted-function list are maintained in ChangeLog.txt.

Roadmap

The current OpenCV 5.0.0 development steps are maintained in ROADMAP.md. Steps are planning and progress markers, not package versions or tags.

Category coverage and priority gaps are tracked in CAPABILITIES.md.

License

GPL. See the repository license and source headers for details.

Navigation Menu

Repository files navigation

IPCV

Highlights

Current Development Status

Codex And GPT-5.6 During OpenAI Build Week

Coordinate Convention

OpenCV Example Assets

Requirements

Naming And Archived Aliases

Versioning

Load IPCV

Build From Source

1. Prepare Compiler Tools

2. Build OpenCV 5 Thirdparty

Windows

Linux And macOS

3. Start Scilab From The Compiler Environment

4. Build IPCV

5. Load And Smoke Test

Graphical Tools

Image And Video Tool

OpenCV Zoo Model Browser

DNN Example

Tests

Repository Layout

Change History

Roadmap

License

About

Releases

Packages

Used by

Contributors

Languages

Name		Name	Last commit message	Last commit date
demos		demos
etc		etc
help		help
images		images
inline/en_US		inline/en_US
macros		macros
sci_gateway		sci_gateway
src		src
tcl		tcl
tests		tests
thirdparty/build		thirdparty/build
.gitattributes		.gitattributes
.gitignore		.gitignore
CAPABILITIES.md		CAPABILITIES.md
ChangeLog.txt		ChangeLog.txt
DESCRIPTION		DESCRIPTION
LICENSE		LICENSE
README.MD		README.MD
ROADMAP.md		ROADMAP.md
VERSION		VERSION
builder.sce		builder.sce
license.txt		license.txt
makedist.sce		makedist.sce

Latest commit

History

Folders and files

Repository files navigation

IPCV

Highlights

Current Development Status

Codex And GPT-5.6 During OpenAI Build Week

Coordinate Convention

OpenCV Example Assets

Requirements

Naming And Archived Aliases

Versioning

Load IPCV

Build From Source

1. Prepare Compiler Tools

2. Build OpenCV 5 Thirdparty

Windows

Linux And macOS

3. Start Scilab From The Compiler Environment

4. Build IPCV

5. Load And Smoke Test

Graphical Tools

Image And Video Tool

OpenCV Zoo Model Browser

DNN Example

Tests

Repository Layout

Change History

Roadmap

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages