Pydoop-features is a suite of tools for extracting features from image data. It uses Bio-Formats to read image data, Avro for (de)serialization and WND-CHARM for feature calculation.
The fastest way to get a working installation is to pull the Docker image:
docker pull imagedata/pyfeatures
Java-Python interoperability is achieved via Avro. The input dataset
can be in any format supported by
Bio-Formats. For
instance, download
MF-2CH-Z-T
and unpack it under /tmp. The first step is to serialize this data
to Avro:
docker run -u ${UID} --rm -v /tmp:/tmp imagedata/pyfeatures \
serialize /tmp/MF-2CH-Z-T.tif -o /tmp/
You should get one avro container file per image series in the input dataset. In this case:
/tmp/MF-2CH-Z-T_{0,1,2,3,4}.avro
To compute features for the first avro container:
docker run -u ${UID} --rm -v /tmp:/tmp imagedata/pyfeatures \
calc /tmp/MF-2CH-Z-T_0.avro -o /tmp/
You might want to get a cup of coffee, feature calculation takes time.
When the above finishes, you should have the following file:
/tmp/MF-2CH-Z-T_0_features.avro
which can be read from either Java or Python. For instance:
>>> from avro.datafile import DataFileReader
>>> from avro.io import DatumReader, BinaryDecoder
>>> with open("/tmp/MF-2CH-Z-T_0_features.avro") as f:
... reader = DataFileReader(f, DatumReader())
... records = [_ for _ in reader]
...
>>> len(records)
40
>>> r = records[0]
>>> r['haralick_textures']
[0.0015474594757607179, 0.00029323128834782644, ...]