Multi-View Linear Discriminant Analysis Network.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Real-world objects are often described from multiple perspectives, necessitating multi-view analysis.
• Significant view discrepancy poses challenges for cross-view recognition and retrieval tasks.

Purpose of the Study:

• To develop a novel method, the Multi-view Linear Discriminant Analysis Network (MvLDAN), for learning a nonlinear, view-invariant representation across multiple views.
• To enhance cross-view recognition and retrieval by minimizing view discrepancy.

Main Methods:

• The MvLDAN utilizes multiple feedforward neural networks, one for each view.
• A novel eigenvalue-based multi-view objective function is proposed to maximize discriminative variance within common feature dimensions.
• The objective function ensures low intra-class variance, high inter-class variance, and high covariance between views.

Main Results:

• The MvLDAN generates deep features that can be projected into a latent common space.
• In this space, samples from the same class are clustered closely, irrespective of their original view.
• Samples from different classes are separated effectively, regardless of their view.

Conclusions:

• The proposed MvLDAN effectively addresses view discrepancy in multi-view analysis.
• Experimental validation on five databases demonstrates superior performance compared to 19 state-of-the-art methods.