Efficient robot object recognition technique based on distance Kernel PCA

Abstract

Feature extraction is the key issue in a recognition system. Principal Component Analysis (PCA) is one of the most widely used feature extraction algorithms. But it is inadequate for this linear method to describe real images which contain complex nonlinear variations, such as illumination, distortion and so on. In this paper, an efficient object recognition method based on distance Kernel PCA (KPCA) is proposed. First, a new kernel called distance kernel is presented to set up the corresponding relation between the higher-dimensional feature space and the original input space. Then, PCA was performed in the higher-dimensional space and a nearest neighbor strategy was used for decision-making. The experiments on both ORL face database and general object image dataset collected by the robot camera illustrate that KPCA with the distance kernel outperforms PCA in robot object recognition: higher recognition accuracy and less computing time.