Small_sample_recognition
Small sample computer motherboard recognition.
Principle
The model is based on the sliding window BOF algorithm.
Algorithm framework
- Feature extraction: Use SIFT for feature extraction.
- Vocabulary construction: K-means is used for clustering. Follow-up tests show that taking 1024 classes works best.
- Category judgment: After clustering, we get Bag of visual words histograms. Use SVM as discriminative classifier.
- Generate bounding box: Reduce the sliding step and window size, and merge the small bounding boxes into larger bounding boxes to make the windows more accurate and reasonable in size.