This is a competition hosted on Kaggle, which was completed as my final project for Machine Learning course. The main purpose of this project is to recognize the breed of a Dog using different machine learning techniques. The scope of project was reduced to identifying just 6 breeds instead of 120 breeds, namely boston bull, chihuahua, doberman, golden retriever, irish wolfhound and redbone.
Every dog have some features which are innate to a breed. Using those features we can train a machine to correctly predict a dog's breed.
The first step was to extract various features from the training image by using the SIFT algorithm. After extracting the features, we use the K-means clustering algorithm to group together similar kind of features which provide an approximate estimate as to what the image is. Using the result from K-means clustering, bag of features is created which contains the frequency of features belonging to different cluster for a given image. Using bag of features, different classifiers were trained (in our case KNN, SVM and AdaBoost).
The program was tested for 6 different values of k. However, the most optimum result for each classifier was obtained by using different values of the clusters. i.e. for k= 10, 20, 30, 40, 50, 60. Below is a table for results obtained using different values for k.
| k | KNN | SVM | AdaBoost |
|---|---|---|---|
| 10 | 44.27 | 31.14 | 30.93 |
| 20 | 47.24 | 40.25 | 40.46 |
| 30 | 50.21 | 38.93 | 39.83 |
| 40 | 48.51 | 44.06 | 45.33 |
| 50 | 49.36 | 42.16 | 44.06 |
| 60 | 49.36 | 43.85 | 43.00 |
Competition Page: https://www.kaggle.com/c/dog-breed-identification/data
Stanford Dog Data Set: http://vision.stanford.edu/aditya86/ImageNetDogs/