PyVision is a object-oriented Computer Vision Toolkit (BSD Licenses)
that is designed to allow the rapid prototype and analysis of computer
vision algorithms. Using python as a foundation this package provides
a simple framework that unifies the Matlab like functionality of
scipy/numpy, Open Computer Vision Library (opencv), and other vision
and machine learning software packages. In addition PyVision provides
a set of analysis tools that allows the researcher to evaluate there
algorithms in python or to export those results in csv format for
analysis in Excel, R, or SAS.
PyVision is a collection of computer vision algorithms that I have implemented
as part of my Ph.D. face recognition research. The purpose of the library is
to provide a set of utilities that interface with common computer vision and
machine learning libraries that will allow rapid vision algorithm prototyping.
For example the current implementation has an image class which will easily
translate image data between PIL, numpy array, and opencv formats.
Currently, PyVision contains a set of popular computer vision algorithms
* canny edge detection (opencv)
* harris corner detection (opencv)
* viola and jones face detection (opencv)
* support vector machines (libsvm)
* image normalization primitives (scipy)
* Difference of Gaussian ROI detection (scipy)
* Genetic Algorithm (python)
* Phase Correlation (scipy)
* Eye Detection (libsvm/scipy)
* PCA analysis (scipy)
* PCA Face Recognition (scipy)
* Image Affine Transform (scipy)
Pyvision also has a set of built in analysis tools (based on scipy) for
analyzing and testing vision algorithms. This includes a simple image
annotation frame work formal statistical analysis that produces confidence
limits and p-values, classes designed for face detection and face recognition
tests, and a table class which allows experiment data to be accumulated and
then exported to csv, txt, or tex formats. PyVision also has a simple GUI
(based on opencv and wxpython) which allows for the easy creation of live
My goals for releasing this library on sourceforge is to provide this
toolkit to other computer vision researchers. I would also like to have a
place where other researchers contribute their code, and I could use some
help making the toolkit multi-platform. The current version is focused on
face recognition, and I would like some help expanding the library to other
areas of computer vision and machine learning.
This library is a set of useful python classes for quickly constructing
and analyze computer vision systems. The library provides a set of
easy to uses object and algorithms that can be used to quickly prototype
new vision algorithms.
This library contains vision algorithms that should be of general use
to many people. As best as possible I will include training configurations,
sample code, and unit tests to aid others in reusing the code.
My current research is focused on face recognition so much of the active
development will be related that area. Contributions in other areas of
research are welcome including: Object Recognition, Biological Vision,
Multiview Geometry, etc.
Python based for rapid prototyping.
Common data types with easy conversion.
Automatically tuned or easy to configure algorithms.
Numerical support from scipy.
Image support from PIL.
Pretrained algorithms for common tasks.
Python and R based algorithms analysis.
Unit tests for algorithm verification.
Sample code and tutorials.
Save and load configurations with python's pickle.
Simple interfaces to third party software: OpenCV and libsvm.
Image.asOpenCV() - Requires fix to opencv bug #1618474 included in cvs on
Jan 25, 2007 18:19:21 UTC by rstanchak
1.0.0 - Many interface things have been cleaned up.
Function arguments that are keywords have been change and may not
work as in previous version. These include: buffer, type, filter, iter
All warnings and errors from the code analysis have been fixed.
0.8.1 - Released in July 2010. This is the last release before a major
upgrade and the last version to support OpenCV 1.0 with the swig
0.8.0 - Released in June 2009. This was primarily to support FaceL.