Dimensionality reduction through LDA and bag-of-features applied to image retrieval

Although the application of dimensionality reduction techniques is promising, the required computational time to calculate the matches between every image through the reduced feature vectors is still considerable. Therefore, the use of bag-of-features becomes a suitable solution since it preserves the individual influence of the feature vectors while adding generalization (Csurka, Dance, F., Willamowski, & Bray 2004) and has demonstrated to achieve good performance (Lazebnik, Schmid, & Ponce 2006, Zhang, Tian, Hua, Huang, & Li 2009, Jurie & Triggs 2005).