Two pass K-means algorithm for finding SIFT clusters in an image
This paper explores the ways to represent images as bags of SIFT feature clusters. SIFT features themselves are widely used in image analysis because of their properties of scale and rotation invariance. The usual way to group them is to segment the image into regions and then assign features to the corresponding image parts. When images themselves are not available for privacy reasons, this is not possible. We created a hybrid clustering algorithm which offers more flexibility than simple spatial k-means clustering. The algorithm parameters were optimized by a stochastic procedure. The impact of different elements of local representation on final clustering quality is discussed.