Automatic flower recognition
The objective with this project was to design and implement a MATLAB program to automatically classify flower
images. Given both train/test sets we had to select the classifier and the features that resulted in better
empirical results.
The first part of the project consisted in choosing the classifier and features that provided the most accurate
result (note: neural networks were not allowed for this project). After some experimentation with the test set
that was already segmented (more of this in the second part) and our knowledge in the matter, we decided to use
a support-vector machine (SVM) as our classifier. We also decided to use the following flower features for the
classification:
- Mean of the flower color
- Standard deviation of the flower color
- Compactness of the image
- Number of vertices
- Histogram of oriented gradients (HOG)
The second part consisted in obtaining the segmented flower from each image in the test set that wasn't already
segmented, we chose to use automatic segmentation by applying different MATLAB procedures. This is done by obtaining
the threshold of the image and with this value there's a procedure that should obtain the countourn of the flower.
We then improve the countourn and apply it to the original image as a mask.
Finally, the following table shows the values of accuracy obtained by our project:
| Mean accuracy (1) | Mean accuracy (2) | |
|---|---|---|
| Mean of the flower color | 0.2877 | 0.1678 |
| +Standard deviation of the flower color | 0.4969 | 0.2815 |
| +Compactness of the image | 0.6148 | 0.3215 |
| +Number of vertices | 0.6767 | 0.4107 |
| +HOG | 0.7804 | 0.5904 |