Novel Robust and Invariant Feature Extraction by Spatio-Temporal Decomposition of Images

Feature extraction is a major step in all pattern recognition and image processing applications. Conventional feature extraction methods when used for extracting physical quantities like mean, entropy etc. are not suitable for automation due to complexity of the feature extraction process. In this paper we propose a simple and novel feature extraction technique that decomposes the original image into a series of sparse images using a time varying selection criterion on the spatial plane. Features are then extracted from each of these sparse images. The feature set, when carefully analyzed and interpreted, is seen to perform as well or even better than their conventional counterparts for recognition and classification. The technique is demonstrated to be robust against noise and results in highly discriminatory features. Also, in this paper the technique to obtain shift invariant features is proposed.