Dimensionality reduction of SIFT using PCA for object categorization
Nattachai Watcharapinchai, Supavadee Aramvith, Supakorn Siddhichai, Sanparith Marukatat · 2009
The problem of automatic object categorization is investigated under the proposed bag of feature object categorization framework. The framework consists of feature detection and representation which uses the scale invariant feature transform (SIFT) as local feature and bag of feature model to represent the image. Learning process utilizes k-NN (k-nearest neighbour). In this paper, we propose the dimensionality reduction of SIFT using principal component analysis (PCA) on each object category to reduce computational complexity and memory requirement during training process. Experimental results show that our proposed technique can reduce the dimension of SIFT up to around 80% with the same average precision compared to baseline technique without our proposed method.