Bidirectional PCA with assembled matrix distance metric for image recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Principal Component Analysis (PCA) is widely used in image recognition.
• Current research focuses on PCA for feature extraction and classification separately.
• A need exists for integrated methods addressing both aspects simultaneously.

Purpose of the Study:

• To propose a novel method combining feature extraction and classification for image recognition.
• To introduce Bidirectional PCA (BD-PCA) for enhanced dimensionality reduction.
• To present an Assembled Matrix Distance (AMD) metric for improved classification.

Main Methods:

• Utilized Bidirectional PCA (BD-PCA) for feature extraction by reducing dimensionality in both row and column directions.
• Developed an Assembled Matrix Distance (AMD) metric to compute distances between feature matrices.
• Employed nearest neighbor and nearest feature line classifiers in conjunction with BD-PCA and AMD.

Main Results:

• BD-PCA effectively extracts features by reducing data dimensionality.
• The AMD metric accurately calculates distances between feature matrices.
• Experimental results demonstrate the efficiency of the combined BD-PCA and AMD approach.

Conclusions:

• The proposed BD-PCA with AMD metric offers an efficient solution for image recognition.
• Simultaneously addressing feature extraction and classification improves performance.
• This integrated method shows significant potential for advancing image recognition technologies.