Sequential Recognition Systems with Information Feedback.
SOUTHEASTERN MASSACHUSETTS UNIV NORTH DARTMOUTH DEPT OF ELECTRICAL ENGINEERING
Pagination or Media Count:
In the paper, the author examines a class of sequential recognition systems which provide the receptor with information feedback. The informations include context, decisions made and the features already selected. The information feedback implies an on-line process in both feature selection and contextual analysis. The complete recognition system is a totally integrated system which takes into consideration the interactions dependence among all elements of the system. The problem of the overall system optimization is considered in the paper. By examining the average risk function, it is shown that the optimization can in fact be accomplished in two separate steps 1 selecting the best feature set using all available measurements and the contextual information and 2 making the minimum risk decision using the context. The contextual information is based on what is best available. Further study is being made to improve the context with feature extraction. The information feedback can simplify the feature selection process as well as the overall system design to provide the same performance as a much more complex recognition system without feedback. Author