Joint Tensor Feature Analysis For Visual Object Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Tensor-based object recognition is a growing field.
• Existing methods often struggle with joint feature selection from complex tensor data.

Purpose of the Study:

• To propose a novel method, Joint Tensor Feature Analysis (JTFA), for joint feature selection and tensor feature extraction.
• To enhance the performance of tensor-based object recognition.

Main Methods:

• Defined modified within-class and between-class tensor scatter values for regression.
• Combined k-mode optimization with L(2,1)-norm jointly sparse regression for optimal solutions.
• Analyzed convergence, computational complexity, and the method's essence.

Main Results:

• JTFA achieves jointly sparse projections for effective tensor feature extraction.
• The method demonstrates similarity to sparse-constrained SVD and eigen-decomposition.
• Experimental results show JTFA outperforms established tensor feature extraction algorithms.

Conclusions:

• JTFA offers a robust and effective approach for tensor feature selection and recognition.
• The proposed method provides a significant advancement in tensor-based object recognition.
• JTFA's performance validates its utility on various tensor datasets.