AtomSpace Computer Vision Types

Atomese wrappers around a subset of OpenCV

The general idea is that the ImageValues will hold a handle to the image being processed, and the ImageBlurLink, and other things of that kind specify the kinds of transformations to be applied to the image. The actual images are never in the AtomSpace, only the abstract processing tree, and a way of hooking that tree to the image(s) to be processed.

Status

• It is possible to read an image from a file using ImageNode, its name corresponds to the image file location.
• There are a few OpenCV operations that you can invoke both on image Nodes and Values. They return ImageValues. Look down for usage info in Scheme.
• Saving images supported with ImageWriteLink.

Building

Dependencies

• CMake (build dependency)
• OpenCV (libopencv-dev package on Debian-derived Linux distributions)
• AtomSpace

Commands

The basic command is just CMake:

cmake -B build/ -S .

Then cmake --build build and sudo cmake --install build

If you have VCPkg and / or you want to use something like clangd that requires compile_commands.json, there is the configure.sh script that runs a slightly extended version of the above command (export VCPKG_ROOT with the path to your VCPkg).

Library Usage

CMake

• find_package library name: AtomSpaceVision.
• Target to link with: AtomSpace::Vision.

Types Usage

For Scheme users:

• (Image "<file-path>") opens an image from a given file.
• (ImageBlur <Image or ValueOf> (Number <kernel size>)) to setup a cv::blur operation.
• (ImageWrite <Image or ValueOf> <Concept or ValueOf>) writes an image to a given location.
• (ImageHalfSize <Image or ValueOf>) to setup a cv::pyrDown operation.
• (ImageFilter2D <Image or ValueOf> (Number <ddepth>) <Image or ValueOf>:kernel) to setup a cv::filter2D operation.

More plans and ideas

Other required and optional ImageFilterLinks. These are described at https://docs.opencv.org/4.5.3/d4/d86/group__imgproc__filter.html

• display to screen for debugging. So, ImageDisplayLink.

• thresholding, convert to greyscale, convert to binary b/w.

• ImageRectangleLink - this needs to be a white rectangle on a black background, with a given width, height, postion. To be used with the above to select regions (but Rect is a separate type in OpenCV that is used by other procedures for selecting ROI, is this necessary as a Link then?).

• HaarWaveletLink - we want to search for features in a position-independent, scale-independent way. We need a good API for that. Not at all clear what a good API for that might be ... Really, we want something affine... Maybe this: http://wavelet2d.sourceforge.net/

• erode() ... maybe useful later??

• Extract hue, saturation, value, extract red, green blue channels. How to do this in opencv ??

• Maybe a circle or ellipse? to answer yes/no to circular feature questions.