Learning apparatus, a learning system, learning method and a learning program for object discrimination

A learning method comprising:

• generating a weak discriminator which calculates a discrimination score that indicates whether a target instance is a positive instance or not, based on a feature extracted from a plurality of bags and a bag label which is the information whether each bag is a positive bag or negative bag,
• calculating the discrimination score based on the generated weak discriminator,
• calculating a probability (instance probability) that a target instance is an instance (positive instance) of the correct target object based on the calculated discrimination score,
• calculating a probability (bag probability) that no smaller than two positive instances are included in the bag based on the calculated instance probability, and
• calculating the likelihood representing plausibility of the calculated bag probability based on the bag label.

A computer-readable storage medium storing a learning program to make a computer executes a processing, comprising:

• generating a weak discriminator which calculates a discrimination score that indicates whether a target instance is a positive instance or not, based on a feature extracted from a plurality of bags and a bag label which is information whether each bag is a positive bag or negative bag,
• calculating the discrimination score based on the generated weak discriminator,
• calculating a probability (instance probability) that a target instance is an instance (positive instance) of a correct target object based on the calculated discrimination score,
• calculating a probability (bag probability) that no smaller than two positive instances are included in the bag based on the calculated instance probability, and
• calculating the likelihood representing plausibility of the calculated bag probability based on the bag label.