Ref: method for finding optimal k
- Calculated all the WSS errors and their corresponding k values and stored it in dictionary and then choose the k for which there is a sharp turn (Elbow method).
Img1: image showing plot from elbow method - Stored nearest descriptor’s index in
nearest_features.txt
, so to visualize key points, we can access image number and it’s corresponding key point and descriptor for visualization.
- use TF-IDF for histogram matching
- instead of randomly intializing cluster-centroids, use K-means++ algorithm.