Project Video
Watch the full project presentation and demo on YouTube.
Interactive Demo
Generate textures and watch real-time feature extraction: edges, frequency spectrum, HOG, and classification.
About This Project
This project develops a comprehensive texture image analysis and classification pipeline that extracts multi-scale features from texture images and classifies them using machine learning. The system combines classical image processing techniques -- Sobel edge detection, Gabor filters, co-occurrence matrices, and histogram of oriented gradients (HOG) -- with modern deep feature extractors to achieve robust texture categorization.
The interactive demo above generates six distinct procedural texture types and performs real-time feature extraction. The edge detection panel highlights structural boundaries using gradient magnitude computation. The frequency analysis panel simulates a 2D FFT power spectrum revealing periodic structure. The HOG visualization shows local gradient orientation distributions, and the classification confidence bars display the model's predictions across texture categories.
Edge Detection
Sobel and Canny edge operators extract structural boundaries and gradient magnitudes for texture boundary characterization.
Frequency Analysis
2D FFT power spectrum reveals periodic structure, dominant spatial frequencies, and directional energy distribution.
HOG Features
Histogram of oriented gradients captures local shape and edge orientation statistics for discriminative texture description.
ML Classification
Multi-class SVM and random forest classifiers trained on extracted feature vectors for real-time texture categorization.