BOUNDARY SEGMENTATION

---

About

Superpixel based SNIC Implementation for image segmentation and boundary delineation. This is a wrapper for Simple Non-Iterative Clustering algorithm. Use this for segementing different areas of image in unsupervised manner, based on spectral differences

Parameters

• rasterio_image_object : rasterio object

  Rasterio object is required for geo-referencing of the output. Just pass in rasterio.open() here

• seed : int, eg: 400

  Also known as number of segments in your image. Larger values give more closed polygons, there is no optimal value. Eg: If you give 100, it will try to create 100 closed polygons

• output_file_name: string, eg: 'example.geojson'

  Specify the filename/full path where you want to output the geojson file

Returns :

• Function returns:

  1. File Name
  2. GeoJson File

DETECTING BOUNDARIES

Superpixel: An un-supervised algorithm that works by planting "seeds" across the image, which is required as an input. The idea is to cluster areas by pixel similarity of neighbors of the seeds. Can be very sensitive to pixel changes. The seed values is directly proportional to the number of areas to segment.

• High seed values will give you more number of closed polygons
• High compactness values will give more uniform boundaries, more squared, but will pick up less variations

VISUAL RESULTS

• Original Image

• SuperPixel Boundary

INSTALLATION

pip install git+https://github.com/purijs/Sentinel-Training-Data-Generation.git

EXTERNAL DEPENDENCIES

gdal >= 3.1

Input image should have at least 12 Bands (Sentinel - 2A) 

HOW TO RUN

from training.unsupervised.segmentation import Segmentation
import rasterio

obj = Segmentation(rasterio.open('data/myimage.tif'), 400, 'sample.geojson')
obj.exportGeoJsonFile()

Check out the example file [here]

Download sample sentinel file here