117 lines
4 KiB
Python
117 lines
4 KiB
Python
"""Experiment on processing KOSMOS data using MorphoCut."""
|
|
|
|
import datetime
|
|
import os
|
|
|
|
from skimage.util import img_as_ubyte
|
|
|
|
from morphocut import Call
|
|
from morphocut.contrib.ecotaxa import EcotaxaWriter
|
|
from morphocut.contrib.zooprocess import CalculateZooProcessFeatures
|
|
from morphocut.core import Pipeline
|
|
from morphocut.file import Find
|
|
from morphocut.image import (
|
|
ExtractROI,
|
|
FindRegions,
|
|
ImageReader,
|
|
ImageWriter,
|
|
RescaleIntensity,
|
|
RGB2Gray,
|
|
)
|
|
from morphocut.stat import RunningMedian
|
|
from morphocut.str import Format
|
|
from morphocut.stream import TQDM, Enumerate
|
|
|
|
# import_path = "/data-ssd/mschroeder/Datasets/Pyrocystis_noctiluca/RAW"
|
|
import_path = "/home/pi/Desktop/PlanktonScope_acquisition/01_16_2020/afternoon/14_1"
|
|
export_path = "/home/pi/Desktop/PlanktonScope_acquisition/01_16_2020/14_1_export"
|
|
archive_fn = os.path.join(export_path, "14_1_morphocut_processed.zip")
|
|
|
|
# Meta data that is added to every object
|
|
global_metadata = {
|
|
"acq_instrument": "Planktoscope",
|
|
"process_datetime": datetime.datetime.now(),
|
|
}
|
|
|
|
if __name__ == "__main__":
|
|
print("Processing images under {}...".format(import_path))
|
|
|
|
# Create export_path in case it doesn't exist
|
|
os.makedirs(export_path, exist_ok=True)
|
|
|
|
# Define processing pipeline
|
|
with Pipeline() as p:
|
|
# Recursively find .jpg files in import_path.
|
|
# Sort to get consective frames.
|
|
abs_path = Find(import_path, [".jpg"], sort=True, verbose=True)
|
|
|
|
# Extract name from abs_path
|
|
name = Call(lambda p: os.path.splitext(os.path.basename(p))[0], abs_path)
|
|
|
|
# Read image
|
|
img = ImageReader(abs_path)
|
|
|
|
# Apply running median to approximate the background image
|
|
flat_field = RunningMedian(img, 10)
|
|
|
|
# Correct image
|
|
img = img / flat_field
|
|
|
|
# Rescale intensities and convert to uint8 to speed up calculations
|
|
img = RescaleIntensity(img, in_range=(0, 1.1), dtype="uint8")
|
|
|
|
# Show progress bar for frames
|
|
TQDM(Format("Frame {name}", name=name))
|
|
|
|
# Convert image to uint8 gray
|
|
img_gray = RGB2Gray(img)
|
|
img_gray = Call(img_as_ubyte, img_gray)
|
|
|
|
# Apply threshold find objects
|
|
threshold = 204 # Call(skimage.filters.threshold_otsu, img_gray)
|
|
mask = img_gray < threshold
|
|
|
|
# Write corrected frames
|
|
frame_fn = Format(os.path.join(export_path, "{name}.jpg"), name=name)
|
|
ImageWriter(frame_fn, img)
|
|
|
|
# Find objects
|
|
regionprops = FindRegions(
|
|
mask, img_gray, min_area=100, padding=10, warn_empty=name
|
|
)
|
|
|
|
# For an object, extract a vignette/ROI from the image
|
|
roi_orig = ExtractROI(img, regionprops, bg_color=255)
|
|
roi_gray = ExtractROI(img_gray, regionprops, bg_color=255)
|
|
|
|
# Generate an object identifier
|
|
i = Enumerate()
|
|
object_id = Format("{name}_{i:d}", name=name, i=i)
|
|
|
|
# Calculate features. The calculated features are added to the global_metadata.
|
|
# Returns a Variable representing a dict for every object in the stream.
|
|
meta = CalculateZooProcessFeatures(
|
|
regionprops, prefix="object_", meta=global_metadata
|
|
)
|
|
# If CalculateZooProcessFeatures is not used, we need to copy global_metadata into the stream:
|
|
# meta = Call(lambda: global_metadata.copy())
|
|
# https://github.com/morphocut/morphocut/issues/51
|
|
|
|
# Add object_id to the metadata dictionary
|
|
meta["object_id"] = object_id
|
|
|
|
# Generate object filenames
|
|
orig_fn = Format("{object_id}.jpg", object_id=object_id)
|
|
gray_fn = Format("{object_id}-gray.jpg", object_id=object_id)
|
|
|
|
# Write objects to an EcoTaxa archive:
|
|
# roi image in original color, roi image in grayscale, metadata associated with each object
|
|
EcotaxaWriter(archive_fn, [(orig_fn, roi_orig), (gray_fn, roi_gray)], meta)
|
|
|
|
# Progress bar for objects
|
|
TQDM(Format("Object {object_id}", object_id=object_id))
|
|
|
|
# Execute pipeline
|
|
p.run()
|
|
|