Scaleable object recognition with a belief model
First Claim
1. A computer readable medium encoded with instructions for causing a computer to recognize objects in content by following the instructions to implement:
- a blackboard comprisinga plurality of experts adapted to process data on a computer, and said data comprising at least one of original input data and/or data created by processing of any of said plurality of experts, anda controller operative to control said experts;
a belief model, coupled to said controller, comprising a set of beliefs and probabilities associated with each belief of said set of beliefs, wherein said belief model comprises a set of rules deduced from a computer-implemented learning system, said learning system comprising truth data files for deducing said set of beliefs, said probabilities and shadow objects, a learning system controller and a statistics space controlled by said learning system controller, wherein said set of rules describes how different classes recognized by said learning system are related to each other spatially and physically;
a belief network, coupled to said controller; and
a relations subsystem, coupled to said controller.
8 Assignments
0 Petitions
Accused Products
Abstract
An exemplary embodiment of the present invention is directed to a system, method and computer program product for providing an object recognition blackboard system architecture. The system for recognizing objects in content can include: a blackboard comprising a plurality of experts, and data comprising original input data and data created by processing of any of the plurality of experts, and a controller operative to control the experts; a belief model, coupled to the controller, comprising a set of beliefs and probabilities associated with each belief of the set of beliefs; a belief network, coupled to the controller; and a relations subsystem, coupled to the controller.
-
Citations
20 Claims
-
1. A computer readable medium encoded with instructions for causing a computer to recognize objects in content by following the instructions to implement:
-
a blackboard comprising a plurality of experts adapted to process data on a computer, and said data comprising at least one of original input data and/or data created by processing of any of said plurality of experts, and a controller operative to control said experts; a belief model, coupled to said controller, comprising a set of beliefs and probabilities associated with each belief of said set of beliefs, wherein said belief model comprises a set of rules deduced from a computer-implemented learning system, said learning system comprising truth data files for deducing said set of beliefs, said probabilities and shadow objects, a learning system controller and a statistics space controlled by said learning system controller, wherein said set of rules describes how different classes recognized by said learning system are related to each other spatially and physically; a belief network, coupled to said controller; and a relations subsystem, coupled to said controller. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
-
-
19. A computer-readable medium encoded with instructions for causing a computer to implement a method of recognizing objects comprising:
-
identifying classes of objects specified by a user using a plurality of cooperative object recognition experts; achieving higher accuracy from using in parallel said plurality of cooperative object recognition experts than is achievable using in serial said plurality of cooperative object recognition experts; supporting scaleability of performance including supporting multiple computer processors; developing a belief model by deducing a set of rules from a computer-implemented learning system, said learning system comprising truth data files for deducing beliefs, probabilities and shadow objects, a learning system controller and a statistics space controlled by said learning system controller, said set of rules describing how different classes recognized by said learning system are related to each other spatially and physically, the developing of said belief model including specifying specified associations among said objects, learning learned associations among said objects, representing said specified and learned associations, and forming a belief network wherein said belief network is at least one of a Bayesian Network and a Dempster Shafer Network; and deducing said shadow objects from said belief model.
-
-
20. A computer-readable medium encoded with instructions for causing a computer to implement a method for adding a new expert to a blackboard comprising:
-
creating an expert adapted for processing data on a computer; encapsulating said expert; compiling said expert; adding a stub function to a blackboard; determining if output of said expert is new and if new, then adding the output'"'"'s class to said blackboard, and updating a belief model by providing truth data file data to a computer-implemented learning system, said learning system comprising truth data files for deducing beliefs, probabilities and shadow objects, a learning system controller and a statistics space controlled by said learning system controller; creating a rule to control when said new expert is to be executed when supporting evidence is found to exceed an adaptable threshold; and deducing a set of rules from said learning system, said set of rules describing how different classes recognized by said learning system are related to each other spatially and physically.
-
Specification