Abstract
This thesis investigates a new on-line unsupervised object-feature extraction method that reduces the complexity and costs associated with the analysis of the multispectral image data and the data transmission, storage, archival and distribution as well. Typically in remote sensing a scene is represented by the spatially disjoint pixel-oriented features. It would appear possible to reduce data redundancy by an on-line unsupervised object-feature extraction process, where combined spatial-spectral object's features, rather than the original pixel-features, are used for multispectral scene representation. The ambiguity in the object detection process can be reduced if the spatial dependencies, which exist among the adjacent pixels, are intelligently incorporated into the decision making process. We define the unity relation that must exist among the pixels of an object. The unity relation can be constructed with regard to the: adjacency relation, spectral-feature and spatial-feature characteristics in an object; e.g. AMICA (Automatic Multispectral Image Compaction Algorithm) uses the within object pixel feature gradient vector as a valuable contextual information to construct the object's features, which preserve the class separability information within the data. For on-line object extraction, we introduce the path-hypothesis, and the basic mathematical tools for its realization are introduced in terms of a specific similarity measure and adjacency relation. AMICA is an example of on-line preprocessing algorithm that uses unsupervised object feature extraction to represent the information in the multispectral image data more efficiently. As the data are read into the system sequentially, the algorithm partitions the observation space into an exhaustive set of disjoint objects simultaneously with the data acquisition process, where, pixels belonging to an object form a path-segment in the spectral space. Each path-segment is characterized by an object-feature set. Then, the set of object-features, rather than the original pixel-features, is used for data analysis and data classification. AMICA is applied to several sets of real image data, and the performance and reliability of features is evaluated. Example results show an average compaction coefficient of more than 20/1 (this factor is data dependent). The classification performance is improved slightly by using object-features rather than the original data, and the CPU time required for classification is reduced by a factor of more than 20 as well. The feature extraction process may be implemented in real time, thus the object-feature extraction CPU time is neglectable; however, in the simulated satellite environment the CPU time for this process is less than 15% of CPU time for original data classification.
Date of this Version
8-1-1988
Comments
PhD thesis. The work described in this report was supported in part by NASA Grant NAGW-925.