Graph-Based Image Segmentation using SLIC

A Python implementation of graph-based image segmentation using the SLIC (Simple Linear Iterative Clustering) superpixel algorithm via OpenCV, partitioning images into perceptually meaningful regions based on colour and spatial similarity.

---

Overview

Graph-based image segmentation treats an image as a weighted graph where each pixel is a node and edge weights represent the dissimilarity between neighbouring pixels. The SLIC algorithm clusters pixels into compact superpixels by iteratively minimising a combined colour and spatial distance metric in the CIELAB colour space. Each resulting segment is rendered with a distinct random colour to visualise the partition.

This approach enables robust segmentation across images with diverse intensity distributions, complex textures, and irregular object boundaries.

---

Pipeline

Input Image (RGB)
        │
        ▼
Grayscale Conversion
        │
        ▼
SLIC Superpixel Segmentation
  region_size = k
  ruler = 10.0 (spatial regularisation)
        │
        ▼
Segment Label Extraction
        │
        ▼
Random Colour Assignment per Segment
        │
        ▼
Contour Detection per Segment Mask
(cv2.RETR_EXTERNAL + cv2.CHAIN_APPROX_SIMPLE)
        │
        ▼
Region Filling with Assigned Colours
        │
        ▼
Segmented Output Image

---

Parameters

Parameter	Default	Description
k (region_size)	100	Target superpixel size in pixels — larger values produce fewer, coarser segments
ruler	10.0	Spatial regularisation weight — higher values enforce more uniform superpixel shapes
min_size	50	Minimum superpixel size — segments below this threshold are merged with neighbours
sigma	0.5	Gaussian smoothing applied before segmentation — higher values smooth boundaries

---

Results

Input: High-contrast RGB image with complex foreground and textured background Output: Colour-coded segmentation map separating regions by colour and texture similarity

The SLIC algorithm successfully partitions the image into spatially compact, perceptually uniform superpixels, clearly delineating foreground objects from background regions.

---

Project Structure

graph-image-segmentation-slic/
├── graph_image_segmentation.py   # Segmentation pipeline
└── README.md

---

Requirements

opencv-python>=4.5.0
opencv-contrib-python>=4.5.0
numpy

Install dependencies:

pip install opencv-python opencv-contrib-python numpy

opencv-contrib-python is required for cv2.ximgproc.createSuperpixelSLIC.

---

Usage

1. Place your input image in the project directory
2. Update the image_path variable in graph_image_segmentation.py
3. Run:

python graph_image_segmentation.py

Two windows will open displaying the original image and the segmented output. Press any key to close.

---

Tech Stack

---

References

• Achanta, R. et al. (2012). SLIC Superpixels Compared to State-of-the-Art Superpixel Methods. IEEE Transactions on Pattern Analysis and Machine Intelligence, 34(11), 2274–2282.
• Felzenszwalb, P. & Huttenlocher, D. (2004). Efficient Graph-Based Image Segmentation. International Journal of Computer Vision, 59(2), 167–181.

---

Academic Context

Computer Vision — M.Eng. Information Technology SRH Hochschule Heidelberg, Germany Supervised by Prof. Dr. Milan Gnjatovic