Cauchy Estimator Discriminant Learning for RGB-D Sensor-based Scene Classification

Abstract

Because depth information has shown its effectiveness in scene classification, RGB-D sensor-based scene classification has received wide attention. However, when images are polluted by noise in the transmission process, the recognition rate will decline significantly. Furthermore, after adopting feature representation schemes, the dimensionality of concatenated features that are extracted from the RGB image and depth image pair is very high. Therefore, a new dimensional reduction algorithm called Cauchy estimator discriminant learning (CEDL) is presented in this paper. CEDL simultaneously addresses two goals: (1) to decrease negative influences to some extent when there is noise in the input samples; (2) to preserve the local and global geometry structure of the input samples. Experiments with the frequently used NYU Depth V1 dataset suggest the effectiveness of CEDL compared with other state-of-the-art scene classification methods.