Improved local descriptor (ILD): a novel fusion method in face recognition

Area of Science:

• Computer Vision
• Pattern Recognition
• Machine Learning

Background:

• Feature descriptors are crucial for image recognition.
• Fusing multiple features enhances recognition rates compared to single descriptors.
• Existing methods often rely on single descriptors, limiting performance.

Purpose of the Study:

• To introduce a novel fused local descriptor, the Improved Local Descriptor (ILD).
• To combine the strengths of Local Binary Patterns (LBP), Extended LBP (ELBP), Median Baseline Profile (MBP), and Local Phase Quantization (LPQ).
• To evaluate the performance of ILD against individual descriptors and benchmark methods.

Main Methods:

• Developed ILD by integrating features from LBP, ELBP, MBP, and LPQ into a histogram feature.
• Employed Principal Component Analysis (PCA) for feature compression.
• Utilized Support Vector Machines (SVMs) and Neural Networks (NN) for classification.

Main Results:

• ILD demonstrated superior performance compared to individual descriptors (LBP, ELBP, MBP, LPQ).
• The fused descriptor achieved state-of-the-art results on the ORL, GT, and Faces94 face recognition datasets.
• ILD effectively combined the merits of its constituent descriptors for robust feature extraction.

Conclusions:

• The proposed Improved Local Descriptor (ILD) offers significant improvements in image recognition accuracy.
• Feature fusion is a viable strategy for developing more powerful local descriptors.
• ILD presents a promising alternative for various pattern recognition and computer vision applications.