Neural network for locating and recognizing a deformable object

A system for automatically detecting and recognizing the identity of a deformable object within an arbitrary image scene, said system comprising:

• object detector means for determining whether said object is within said arbitrary image scene;
• feature localizer means for determining the position of an identifying feature on said object, said feature locator means being coupled to said object detector means;
• feature extractor means coupled to said feature localizer, for receiving coordinates sent from said feature localizer which are indicative of the position of said identifying feature and for extracting from said coordinates information relating to other features of said object which is used to create a low resolution image of said object;
• object recognizer means for determining the identity of said object, said object recognizer means being coupled to said feature extractor means and being operative to receive said low resolution image of said object inputted from said feature extractor means to identify said object;
• wherein said object detector means, said feature localizer means, and said object recognizer means are each implemented in a probabilistic decision-based neural network;
• and
• wherein said probabilistic decision-based neural network comprises a plurality of subnets, each of said subnets having a plurality of cluster basis functions which include cluster prior probabilities according to:
• ##EQU18##

A system for automatically detecting and recognizing the identity of a deformable object within an arbitrary image scene, said system comprising:

• object detector means for determining whether said object is within said arbitrary image scene;
• feature localizer means for determining the position of an identifying feature on said object, said feature locator means being coupled to said object detector means;
• feature extractor means coupled to said feature localizer, for receiving coordinates sent from said feature localizer which are indicative of the position of said identifying feature and for extracting from said coordinates information relating to other features of said object which is used to create a low resolution image of said object;
• and
• object recognizer means for determining the identity of said object, said object recognizer means being coupled to said feature extractor means and being operative to receive said low resolution image of said object inputted from said feature extractor means to identify said object;
• wherein said object detector means, said feature localizer means, and said object recognizer means are each implemented in a probabilistic decision-based neural network;
• and
• wherein said probabilistic decision-based neural network comprises a plurality of subnets, each of said subnets includes a plurality of elliptic basis functions according to:
• ##EQU19##

A system for automatically detecting and recognizing the identity of a deformable object within an arbitrary image scene, said system comprising:

• object detector means for determining whether said object is within said arbitrary image scene;
• feature localizer means for determining the position of an identifying feature on said object, said feature locator means being coupled to said object detector means;
• feature extractor means coupled to said feature localizer, for receiving coordinates sent from said feature localizer which are indicative of the position of said identifying feature and for extracting from said coordinates information relating to other features of said object which is used to create a low resolution image of said object;
• and
• object recognizer means for determining the identity of said object, said object recognizer means being coupled to said feature extractor means and being operative to receive said low resolution image of said object inputted from said feature extractor means to identify said object;
• wherein said object detector means, said feature localizer means, and said object recognizer means are each implemented in a probabilistic decision-based neural network;
• and
• wherein said probabilistic decision-based neural network comprises a plurality of subnets, each of said subnets including a discriminant function which comprises a nonlinearity operator.

A system for automatically detecting and recognizing the identity of a deformable object within an arbitrary image scene, said system comprising:

• object detector means for determining whether said object is within said arbitrary image scene;
• feature localizer means for determining the position of an identifying feature on said object, said feature locator means being coupled to said object detector means;
• feature extractor means coupled to said feature localizer, for receiving coordinates sent from said feature localizer which are indicative of the position of said identifying feature and for extracting from said coordinates information relating to other features of said object which is used to create a low resolution image of said object;
• and
• object recognizer means for determining the identity of said object, said object recognizer means being coupled to said feature extractor means and being operative to receive said low resolution image of said object inputted from said feature extractor means to identify said object;
• wherein said object detector means, said feature localizer means, and said object recognizer means are each implemented in a probabilistic decision-based neural network;
• and
• further comprising object verifier means implemented as a probabilistic decision-based neural network for verifying the decision of said object recognizer means, said object verifier receiving additional information about said object which is cascaded in sequential processing stages in a hierarchical manner by said object verifier.

A system for automatically detecting and recognizing the identity of a deformable object within an arbitrary image scene, said system comprising:

• object detector means for determining whether said object is within said arbitrary image scene;
• feature localizer means for determining the position of an identifying feature on said object, said feature locator means being coupled to said object detector means;
• feature extractor means coupled to said feature localizer, for receiving coordinates sent from said feature localizer which are indicative of the position of said identifying feature and for extracting from said coordinates information relating to other features of said object which is used to create a low resolution image of said object;
• and
• object recognizer means for determining the identity of said object, said object recognizer means being coupled to said feature extractor means and being operative to receive said low resolution image of said object inputted from said feature extractor means to identify said object;
• wherein said object detector means, said feature localizer means, and said object recognizer means are each implemented in a probabilistic decision-based neural network;
• and
• wherein said probabilistic decision-based neural network comprises a plurality of classifier channels each having an output, said outputs being laterally fused by weighting said channels.

A system for automatically detecting and recognizing the identity of a deformable object within an arbitrary image scene, said system comprising:

• object detector means for determining whether said object is within said arbitrary image scene;
• feature localizer means for determining the position of an identifying feature on said object, said feature locator means being coupled to said object detector means;
• feature extractor means coupled to said feature localizer, for receiving coordinates sent from said feature localizer which are indicative of the position of said identifying feature and for extracting from said coordinates information relating to other features of said object which is used to create a low resolution image of said object;
• and
• object recognizer means for determining the identity of said object, said object recognizer means being coupled to said feature extractor means and being operative to receive said low resolution image of said object inputted from said feature extractor means to identify said object;
• wherein said object detector means, said feature localizer means, and said object recognizer means are each implemented in a probabilistic decision-based neural network;
• and
• wherein said probabilistic decision-based neural network comprises a plurality of classifier channels each having an output, said outputs being laterally fused by class-dependent channel fusion according to:
• ##EQU21##

A system for automatically detecting and recognizing the identity of a deformable object within an arbitrary image scene, said system comprising:

• object detector means for determining whether said object is within said arbitrary image scene;
• feature localizer means for determining the position of an identifying feature on said object, said feature locator means being coupled to said object detector means;
• feature extractor means coupled to said feature localizer, for receiving coordinates sent from said feature localizer which are indicative of the position of said identifying feature and for extracting from said coordinates information relating to other features of said object which is used to create a low resolution image of said object;
• and
• object recognizer means for determining the identity of said object, said object recognizer means being coupled to said feature extractor means and being operative to receive said low resolution image of said object inputted from said feature extractor means to identify said object;
• wherein said object detector means, said feature localizer means, and said object recognizer means are each implemented in a probabilistic decision-based neural network;
• and
• wherein said probabilistic decision-based neural network comprises a plurality of classifier channels each having an output, said outputs being laterally fused by data-dependent channel fusion according to:
• ##EQU22##

A system for automatically detecting and recognizing the identity of a deformable object within an arbitrary image scene, said system comprising:

• object detector means for determining whether said object is within said arbitrary image scene;
• feature localizer means for determining the position of an identifying feature on said object, said feature localizer means being coupled to said object detector means;
• feature extractor means coupled to said feature localizer, for receiving coordinates sent from said feature localizer which are indicative of the position of said identifying feature and for extracting from said coordinates information relating to other features of said object which is used to create a low resolution image of said object;
• and
• object recognizer means for determining the identity of said object, said object recognizer means being coupled to said feature extractor means and being operative to receive said low resolution image of said object inputted from said feature extractor means to identify said object,
• wherein said object detector means, said feature localizer means, and said object recognizer means are each implemented in a decision-based one-class-in-one neural network structure having a plurality of subnets and a plurality of classes, wherein each one of said subnets is designated to one of said classes in order to distinguish it from said other classes, wherein said neural network includes a training scheme having a first phase and a second phase, wherein said first phase includes individually training each of said subnets without mutually exchanging information between said classes and said second phase includes reinforcing learning and anti-reinforcing learning obtained during said first phase.