How to polygonize raster to shapely polygons

Question

I am seeking an open-source python solution to convert raster to polygon (no ArcPy).

I did know the GDAL function to convert raster to polygon, and here is the manual: http://pcjericks.github.io/py-gdalogr-cookbook/raster_layers.html#polygonize-a-raster-band

Nevertheless, I expect that the output can be shapely polygons or any object, temporarily in memory, not saved as a file. Is there any package or code to handle this issue?

If the raster has been processed in a numpy array, the approach is listed below.

Answer 1

Use rasterio of Sean Gillies. It can be easily combined with Fiona (read and write shapefiles) and shapely of the same author.

In the script rasterio_polygonize.py the beginning is

import rasterio
from rasterio.features import shapes
mask = None
with rasterio.Env():
    with rasterio.open('a_raster') as src:
        image = src.read(1) # first band
        results = (
        {'properties': {'raster_val': v}, 'geometry': s}
        for i, (s, v) 
        in enumerate(
            shapes(image, mask=mask, transform=src.transform)))

The result is a generator of GeoJSON features

 geoms = list(results)
 # first feature
 print geoms[0]
 {'geometry': {'type': 'Polygon', 'coordinates': [[(202086.577, 90534.3504440678), (202086.577, 90498.96207), (202121.96537406777, 90498.96207), (202121.96537406777, 90534.3504440678), (202086.577, 90534.3504440678)]]}, 'properties': {'raster_val': 170.52000427246094}}

That you can transform into shapely geometries

from shapely.geometry import shape
print shape(geoms[0]['geometry'])
POLYGON ((202086.577 90534.35044406779, 202086.577 90498.96206999999, 202121.9653740678 90498.96206999999, 202121.9653740678 90534.35044406779, 202086.577 90534.35044406779))

Create geopandas Dataframe and enable easy to use functionalities of spatial join, plotting, save as geojson, ESRI shapefile etc.

geoms = list(results)
import geopandas as gp
gpd_polygonized_raster  = gp.GeoDataFrame.from_features(geoms)

Answer 2

Here is my implementation.

from osgeo import ogr, gdal, osr
from osgeo.gdalnumeric import *  
from osgeo.gdalconst import * 
import fiona
from shapely.geometry import shape
import rasterio.features

#segimg=glob.glob('Poly.tif')[0]
#src_ds = gdal.Open(segimg, GA_ReadOnly )
#srcband=src_ds.GetRasterBand(1)
#myarray=srcband.ReadAsArray() 
#these lines use gdal to import an image. 'myarray' can be any numpy array

mypoly=[]
for vec in rasterio.features.shapes(myarray):
    mypoly.append(shape(vec))

The way to install rasterio is by 'conda install -c https://conda.anaconda.org/ioos rasterio', if there is an installation issue.

Answer 3

# PYTHON 3

import rasterio.features
from shapely.geometry import shape

# Mask is a numpy array binary mask loaded however needed
maskShape = rasterio.features.shapes(mask.astype('uint8'))
mypoly=[]
for vec in maskShape:
    mypoly.append(vec[0])
print(mypoly)

This is what worked for me.

Answer 4

Incase of pixels with high variability, convert the arrays to integer type. Helps to save on computation time.

with rio.Env():
with rio.open(input_file_path) as depth_image:
    depth_array=depth_image.read(1).astype(int)