Project from Advanced Image Processing subject to classify 5 popular buildings in Poznan using Bag of Words model.
- opencv-contrib-python 3.4.2.16
- scikit-learn 0.22.2
- numpy 1.17.4
This project is used SIFT detector, so it is important to have old version of OpenCV.
The best results on this train dataset was with image with width resized to 400px. Vocabulary contains 44 words and I used linear SVC model to classify buildings.
Scores on data are presented below:
metrics | Train data | Test data |
---|---|---|
accuracy [%] | 98.73 | 88.33 |
precision [%] | 98.75 | 89.14 |
recall [%] | 98.73 | 88.33 |
Confusion matrix of test data: