CHARACTER_AND_OBJECT_RECOGNITION_BASED_ON_GLOBAL_FEATURE_EXTRACTION
Mind Map: Character and Object Recognition Based on Global Feature Extraction
1. Introduction
- Optical Character Recognition (OCR): Recognizes handwritten, irregular, and machine-printed characters.
- Key Tasks in OCR:
- Pre-processing
- Segmentation
- Feature Extraction
- Classification
- Recognition
2. Feature Extraction Methods
2.1 Global Feature Extraction
- Definition: Uses entire image characteristics to extract features.
- Methods:
- Gray Level Co-occurrence Matrix (GLCM): Uses spatial distribution of gray-level values.
- Edge Direction Matrix (EDMS): Captures edge directions but produces a limited number of features.
- Challenges:
- Less discriminative features.
- Higher dimensionality leads to longer processing times.
2.2 Spatial Feature Extraction
- Definition: Focuses on local image characteristics.
- Techniques:
- Robinson Compass Mask: Uses gradient filters in eight directions.
- Strengths: Better for character recognition.
- Limitations: Time-consuming due to high-dimensional data.
3. Proposed Method
- Combination of GLCM and EDMS:
- Aims to improve recognition rates by combining features.
- Feature Selection: Uses gain ratio to reduce feature set size.
- Datasets: License plates, font styles, and large binary images.
4. Experimental Results
- Performance Metrics:
- Character Recognition Accuracy:
- Proposed method: 85.99% with feature selection.
- EDMS: 80.19%, GLCM: 38.84%, Combination without feature selection: 58.78%.
- Object Recognition Accuracy:
- Proposed method: 92.5% accuracy with feature selection.
- Robinson filter (spatial method) outperforms in character recognition (100% accuracy).
- Character Recognition Accuracy:
- Conclusion: Global feature extraction is better for object recognition, while spatial methods are better for character recognition.
5. Applications
- Character Recognition:
- License Plate Recognition (LPR).
- Handwritten text recognition.
- Object Recognition:
- Recognizing binary shapes in images.
- Differentiating between object categories using extracted features.
6. Future Work
- Improvements: Modify feature selection to further enhance recognition rates.
- New Applications: Adapt method for complex character recognition and extend to other domains.
7. Summary
- Goal: Improve OCR by combining global and spatial feature extraction techniques.
- Key Findings: Proposed method shows promise for object recognition with efficient feature extraction.