Scale-invariant pattern recognition using a combined Mellin radial harmonic function and the bidimensional empirical...

Area of Science:

• Image processing
• Pattern recognition
• Signal processing

Background:

• Traditional pattern recognition methods struggle with scale and shift variations.
• Noise robustness is a critical challenge in real-world pattern recognition applications.

Purpose of the Study:

• To develop a novel scale and shift invariant pattern recognition method.
• To enhance discrimination capability and noise robustness.
• To improve the flatness of peak intensity response versus scale change.

Main Methods:

• Combining bidimensional empirical mode decomposition (BEMD) with Mellin radial harmonic decomposition (MRHD).
• Developing a filter for scale and shift invariant pattern recognition.

Main Results:

• The proposed method demonstrates improved discrimination capability and noise robustness.
• Achieved a flat peak intensity response over a large scale range (0.2 to 1).
• Correlation peak intensity variance remained below 20% within the detection range.
• Experimental validation confirmed numerical simulation results, including performance with white noise.

Conclusions:

• The novel BEMD-MRHD method offers a robust solution for scale and shift invariant pattern recognition.
• The method's uniform response across scales simplifies pattern detection using a global threshold.
• Validated effectiveness in both noise-free and noisy conditions.