Improved Segmentation Model for Melanoma Lesion Detection Using Normalized Cross-Correlation-Based k-Means Clustering

Improved Segmentation Model for Melanoma Lesion Detection Using Normalized Cross-Correlation-Based k-Means Clustering

Melanoma is the most dangerous form of skin cancer, and survival rates exceed 90% when detected early — making accurate lesion segmentation a life-saving capability. This project presents an enhanced image segmentation pipeline for dermoscopy images. The method applies normalized cross-correlation (NCC) to measure pixel similarity against lesion reference patterns, which guides a refined k-means clustering process to precisely separate lesion regions from surrounding healthy skin, hair, and noise. This hybrid technique addresses common segmentation failures caused by low contrast, illumination variation, and irregular lesion borders. The extracted lesion masks feed into a classification module that grades malignancy risk using texture, asymmetry, border, and color features following the clinical ABCD rule. Performance is validated on standard dermoscopy benchmarks using IoU, Dice coefficient, sensitivity, and specificity metrics. Implemented in Python with OpenCV, Scikit-image, and Scikit-learn, and presented through an interactive web interface, this project is a powerful medical image segmentation project for computer vision students, dermatology-tech developers, and researchers working on skin cancer detection using machine learning.

Components

Python 3.8+
OpenCV
Scikit-image
Scikit-learn
NumPy
Streamlit web interface

Key Features

• Normalized cross-correlation guided k-means lesion segmentation
• Robust against low contrast, illumination variation, hair and irregular lesion borders
• ABCD-rule malignancy grading using texture, asymmetry, border and color features
• Validated with IoU, Dice coefficient, sensitivity and specificity on dermoscopy benchmarks
• Interactive web interface for upload, segmentation and risk grading

Applications

Dermatology-tech developers, skin cancer screening initiatives, computer vision students and healthcare AI researchers.