Symmetry axis based object recognition under translation, rotation and scaling

Area of Science:

• Computer Vision
• Pattern Recognition
• Machine Learning

Background:

• Invariant object recognition (IOR) is crucial for real-world applications.
• Existing IOR systems often face challenges with simplicity and accuracy.
• Object recognition needs to be robust to transformations like translation, rotation, and scaling.

Purpose of the Study:

• To develop a new, simple, and accurate invariant object recognition system.
• To introduce the Symmetry Axis based Feature Extraction and Recognition (SAFER) approach.
• To improve upon existing invariant object recognition methods.

Main Methods:

• Feature extraction based on object symmetry axes and concentric circle angles.
• Grouping of extracted features by similarity.
• Utilizing multiple neural networks (NNs) for enhanced training and recognition.

Main Results:

• SAFER demonstrated effective feature extraction and recognition capabilities.
• The system achieved good performance on recognizing English characters and diverse shapes.
• Experimental results indicate SAFER's competitive performance against other algorithms.

Conclusions:

• SAFER provides a promising approach for invariant object recognition.
• The method balances simplicity with high recognition accuracy.
• The use of multiple NNs contributes to improved performance.