Analytical study of two feature extraction methods in comparison with deep learning methods for classification of...

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Accurate classification of small metal objects is crucial in various applications.
• Traditional feature extraction methods like HOG and LBP are widely used but require comparison.
• Deep learning methods offer alternative approaches for image classification tasks.

Purpose of the Study:

• To compare the efficiency of Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP) for classifying small metal objects.
• To evaluate the performance of HOG and LBP against deep learning models such as YOLO and Faster R-CNN.
• To determine the optimal feature extraction strategy for this classification task.

Main Methods:

• Feature extraction using Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP).
• Image classification using K-nearest neighbors, Support Vector Machine, and Naïve Bayesian algorithms.
• Performance evaluation through confusion matrix analysis and comparison with deep learning models.

Main Results:

• Histogram of Oriented Gradients (HOG) demonstrated higher efficiency than Local Binary Pattern (LBP).
• A combination of HOG and LBP features outperformed individual methods.
• Performance comparison indicated the effectiveness of traditional methods in specific contexts.

Conclusions:

• HOG is a more efficient feature extraction method than LBP for classifying small metal objects.
• Combining HOG and LBP features enhances classification accuracy.
• Further research can explore hybrid approaches for improved object recognition.