An angle optimized global embedding algorithm

Traditional algorithm of global dimensionality reduction such as PCA, MDS and Isomap, measure the relation of data by distance, this paper gives an angle measurement approach for the relation of data. Based on the theoretic analysis, a novel angle optimized global embedding (AOGE) algorithm is proposed, which measured the relation of data by the angles between the centralized samples and their orthogonal projections. With the optimization of the angles, global dimensionality reduction embedding is realized. Compared with the global algorithms, such as PCA, MDS etc., the proposed algorithm has more effective results for the datasets containing distance irregularity data or noise data. Experiments on facial expression recognition verified the efficiency of the proposed algorithm.