bilde

(B)uffer (I)maging (L)ibrary, (D)irty and (E)xtendable

Python extention (pybilde)

Use in python:

Label Connected Components of a torch tensor

import pybilde
import torch
from matplotlib import pyplot as plt
bw_img = (torch.rand(1000, 1000) > .8).to(torch.uint8)
l_img, n_components = pybilde.label_connected_components(bw_img)
plt.imshow(l_img, cmap='nipy_spectral'))
plt.show()

Install from sdist

• You will need a C++17 compiler
• Probably a Unix environment (eg ubuntu 24.04)
• If pytorch is installed in the current enironment, the module will build the ptbilde submodule otherwise, only the numpy submodule (npbilde) will be built.

pip install bilde

Building binaries:

• Install build dependencies

pip install cibuildwheel

• Build bdist without a CI

CIBW_BUILD="cp313t-manylinux_x86_64 cp313t-manylinux_x86_64 cp312-manylinux_x86_64 cp311-manylinux_x86_64 cp310-manylinux_x86_64 cp39-manylinux_x86_64 cp38-manylinux_x86_64 cp37-manylinux_x86_64 cp36-manylinux_x86_64" CIBW_BEFORE_BUILD="yum install -y boost-devel" cibuildwheel --platform linux --output-dir wheelhouse
CIBW_SKIP="pp310-manylinux_i686 pp39-manylinux_i686 pp38-manylinux_i686 pp37-manylinux_i686 pp36-manylinux_i686 pp310-manylinux_x86_64  pp39-manylinux_x86_64 pp38-manylinux_x86_64  pp37-manylinux_x86_64 pp36-manylinux_x86_64" CIBW_BEFORE_BUILD="yum install -y boost-devel" cibuildwheel --platform linux --output-dir wheelhouse

C++ library

A library for writing procedural computer vision code in C++

• Header only No linking complications! Just copy it in your source tree.

• High C++ template use

• Define algorithms once, use them directly on several back-ends. With a mechanism based on implicit template instantiation, an algorithm is defined as a single function, and will seemlesly run several image containers such as cv::Mat, IplImage, octave arrays, numpy arrays.

Python module

LBP feature extraction

import pybilde
import skimage
img = skimage.data.coins()
histogram = pybilde.lbp_features(img, 8, [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12])

Connected compoent labeling

import pybilde
import skimage
img = skimage.data.coins()
cc_img, nb_components = pybilde.label_connected_components(img>.5)

Demos:

• SRS-LBP feature extractor:

Cross compiling for win32 (static build) and running in debian based linux.

Install dependencies:

sudo apt-get install autoconf automake autopoint bash bison bzip2 flex gettext git g++ gperf intltool libffi-dev libgdk-pixbuf2.0-dev libtool-bin libltdl-dev libssl-dev libxml-parser-perl make openssl p7zip-full patch perl pkg-config python ruby scons sed unzip wget xz-utils g++-multilib libc6-dev-i386
export MXE_ROOT="$HOME/tools/mxe"
mkdir -p "$HOME/tools"
cd "$HOME/tools"
git clone https://github.com/mxe/mxe.git
cd mxe
make opencv boost

Compile:

./compile_static_mxe.sh ./src/lbpFeatures2.cc /tmp/lbp_features2.exe

Get help:

wine /tmp/lbp_features2.exe

Run:

wine /tmp/lbp_features2.exe -i ./sample_data/PICT2466.png > /tmp/features.csv

Compiling in ubuntu:

sudo apt-get install libopencv-dev libhighgui-dev libprotobuf-dev libwebp-dev #  depending on your system you might need other packages
cd src
make lbpFeatures2