Eigenfish

Eigenfish is a Python 3 package for detecting fish in an image sequence.

Requirements

Requires Python 3. To install, run pip3 install -r requirements.txt from the root directory.

Usage

For a detailed functional example, please see example.py. Documentation is available at docs/_build/html/index.html.

Eigenfish must be trained before it is able to classify an images as follows:

import eigenfish

ef = eigenfish.Eigenfish(image_shape)
ef.train(image_matrix, labels)
result = ef.classify(unlabeled_image_matrix)

where:

image_shape is the (height, width) of all images used
image_matrix is matrix with each column a flattened image
labels is a list of labels with labels[i] corresponding to image_matrix[:, i]
unlabeled_image_matrix is the matrix of flattened images to classify

Additionally, ef.cross_validate(labeled_image_matrix, labels) can be called after ef.train(...) to check accuracy of the trained model.

util.py contains the helper function load_img_mat to make loading images easier.

Save/Load

Eigenfish has support for saving a trained classifier and loading it later, through Eigenfish.save(filename) and Eigenfish.load(filename).

Customization

Custom classifiers and preprocessors can be used with Eigenfish by passing classes to the processor and classifier arguments in the constructor Eigenfish(). See process/process.py and classify/classify.py for the defaults.

Documentation

Run make html from the docs/ directory.

Example

Run python3 example.py from the root directory.

Unit Tests

Run python3 test.py from the root directory.

Copyright

Eigenfish is free and open-source software made available under the MIT License. See LICENSE file for details.

References

[1] Candès, E. J., Li, X., Ma, Y., and Wright, J. Robust principal component analysis? Journal of the ACM, 58(3):11:1-11:37, 2011.

[2] Huang, P.X., Boom, B.J., and Fisher, R.B., Underwater live fish recognition using a balance-guaranteed optimized tree, In Computer Vision ACCV 2012, Lecture Notes in Computer Science Volume 7724, pp. 422- 433, Springer Berlin Heidelberg, 2013.

pens/eigenfish