8.2 BoW Classification
In this lecture, we are going to learn about BoW Classification.
Once again, let's see how Classical Image Classification Pipeline looks.
Among 4 steps, today we are going to focus on 'encoder', 'bag of visual words' encoder.
● BoW encoder
In BoW encoder view, we can define 'encode' and 'decode' as below.
⊙ Encode
1. Dictionary Learning
a. Extract features from image
* Regular grid
- Vogel & Schiele, 2003
- Fei-Fei & Perona, 2005
* Interest point detector
- Csurka et al. 2004
- Fei-Fei & Perona, 2005
- Sivic et al. 2005
* Other methods
- Random sampling (Vidal-Naquet & Ullman, 2002)
- Segmentation-based patches (Barnard et al. 2003)
b. Unsupervised Clustering
In Unsupervised Clustering, the answers are not given.
* 4 steps in Unsupervised Clustering
① Select initial centroids at random
② Assign each object to the cluster with the nearest centroid
③ Compute each centroid as the mean of the objects assigned to it (go to 2)
② Assign each object to the cluster with the nearest centroid (again)
Repeat previous 2 steps until no change.
* K-means Clustering
given k:
① Select initial centroids at random.
② Assign each object to the cluster with the nearest centroid.
③ Compute each centroid as the mean of the objects assigned to it.
④ Repeat previous 2 steps until no change.
2. Build Bags-of-Words (BoW) descriptor
After learning a visual dictionary, an image representation is needed, so we have to use a bag-of-words descriptor.
a. Quantize feature
- Associate each feature to nearest 'word' cluster.
b. Build histogram (for each image)
⊙ Decode
1. Train classifier
- Given the bag-of-features representations of images from different classes, train some fancy machine learning model.
2. Classify using BoW descriptor
- Classify using BoW descriptor
In next lecture, we will learn about decoder, optimizer.