Adaptive texture classification with Hartley transform and application to visual inspection

摘要

This paper describes a vision system designed for automatic classification and inspection of textures based on both the Fast 2-D Hartley transform and the learning vector quantization (LVQ) neural networks. We show that a minimal feature setresults in effective texture discrimination. Hartley transform-based feature extraction results in real-time performance. An LVQ neural network maps each feature vector on a single neuron that represents the texture class. The system has been tested on anumber of synthetic and real textures. An example of defect detection and classification using data on woven aluminum wire webs is presented. We report a 95% accuracy in classification attained using the minimum feature set for classifying a set of 6Brodatz textures.