BIFL

basic image feature library

Python module with functions for image feature extraction based on the OpenCV Python bindings.

Requirements

OpenCV (w/ Python bindings: sudo apt-get install python-opencv)

Python Image Library

Installation

pip install https://github.com/johaness/bifl.git

Usage

The features.extract function allows extraction of all available features on different spatial scales -- see the source documentation for details.

The bifl command line script accepts any number of image file names as parameter, runs features.extract on every image, stores results in a pickle file of cvMat matrices, and renders each feature into a PNG image.

The other modules contain a number of support functions useful for working with OpenCV data structures.

Image Features

Image features implemented in pure Python are defined in mods.py:

contrast(inmat, ws = 51)
smooth(inmat, ws = 51)
sobel(inmat, ws = 7)

pyrdown(inmat)
pyrsdown(*inmats)
pyrup(inmat)

zscale(inmat)
equalize(inmat)

add(*mats)
addZ(*mats)
multiply(inmat, value)
addZW(inmats, weights)
addW(inmats, weights)

spatialbias(inmat, biasmat, (x, y), base=1.0, gain=1.0, bias_zero=None)
maxior(inmat, steps=10, inhibition=0.2, radius=90)

Image features can be implemented in C with a minimal wrapper below bifl/cpy/:

# split RGB into RG, BY, Lum, Sat
colorsplit(inimage)
# intrinsic dimensionality
intdim(inmat)

License

BSD License, see LICENSE file

Development sponsored by WhiteMatter Labs GmbH, creators of EyeQuant.