Spatial-Spectral Features-Based Dimensionality Reduction Technique for Robust Multivariate Image Classification

Recent developments in multivariate imaging have led to several applications for in-depth knowledge in the discipline of remote sensing, medical and forensic. The multiband structure of multivariate data results in high dimensionality during feature extraction. Therefore, dimensionality reduction (DR) becomes the key aspect along with efficient feature extraction technique for multivariate image analysis. The conventional projection-based techniques fail to solve the nonlinearity in high-dimensional multivariate images which results in inadequate classification accuracy. Therefore, prior to DR methods, spatial features described by textural descriptors along with spectral features are fused to give robust feature space. Gray level co-occurrence matrix (GLCM) extracts textural features. Rich spatial-spectral feature space is subjected to a dimensionality reduction approach to limit the effect of curse of dimensionality on classification performance. This paper gives insights about the performance of DR techniques on proposed spatial-spectral feature space.