Feature Thresholding in Generalized Approximation Spaces

Recent advances in information sciences are extending classical set theory frontiers into new domains of uncertainty perception, incompleteness, vagueness of knowledge-giving new mathematical approach to development of intelligent information systems. The paper addresses the problem of construction of rough measures in generalized approximation spaces introducing a new method of rough feature thresholding. The algorithm creates rough feature blocks and assigns them image blocks from the block min, avg, max statistics. The algorithm converts data blocks into rough approximations of feature blocks. The introduced solution contributes to the highly precise internal data structure descriptors on one side and constitutes the algorithmic base for rough data analysis entirely embedded in generalized approximation spaces at the same time. The scope of possible applications includes image descriptors, image thresholding, image classifications.