Method and system for feature detection

A system for feature detection with reliability measure, said system comprising:

• one or more processors;
• multiple input units;
• wherein a second set of data determines a first subset of said multiple input units to be clamped to a first set of data;
• wherein in clamping a first input unit of said multiple input units, said one or more processors dynamically set a value of said first input unit in a first memory unit associated with said first input unit to a corresponding value from said first set of data;
• one or more layers of stochastic units, comprising a first stochastic unit;
• wherein said first stochastic unit has one or more links traceable to a second subset of said multiple input units, wherein a local coverage extent associated with said first stochastic unit limits said one or more links;
• wherein said one or more links represent first weighted connections between said first stochastic unit and second units;
• wherein said second units are among said second subset of said multiple input units or from said one or more layers of stochastic units between said first stochastic unit and said multiple input units;
• wherein said one or more processors set a memory unit associated with said first stochastic unit to a first value generated based on weighted sum of values of memory units associated with said second units per said first weighted connections represented by said one or more links;
• wherein said one or more processors detect sub-features associated with a higher layer of said one or more layers of stochastic units, based on sub-features detected by a lower layer of said one or more layers of stochastic units;
• one or more output units;
• wherein said one or more processors set a memory unit associated with said one or more output units to a value indicating a detected feature based on one or more said detected sub-features associated with one or more layers of one or more units of said one or more layers of stochastic units;
• wherein said one or more processors determine an energy measure corresponding to said first set of data, based on factors comprising of said first weighted connections, said one or more layers of stochastic units, and said first set of data clamped to said first subset of said multiple input units;
• wherein in said determining said energy measure, a contribution of said energy measure from said first stochastic unit is reduced by a coverage factor of said first stochastic unit;
• wherein said coverage factor of said first stochastic unit is determined based on a value representing fraction of input units within said second subset of said multiple input units traceable from said first stochastic unit that are also in said first subset of said multiple input units clamped to said first set of data;
• wherein said one or more processors output to a second memory unit a value representing a quality measure associated with said detected feature determined based on said energy measure and a baseline;
• wherein said quality measure indicates a degree of conformity, a degree of confidence, relative probability, or a degree of reliability associated with said detected feature for said first set of data.

A system for feature detection with conformity measure, said system comprising:

• one or more processors;
• multiple input units;
• wherein a second set of data determines a first subset of said multiple input units to be clamped to a first set of data;
• wherein in clamping a first input unit of said multiple input units, said one or more processors dynamically set a value of said first input unit in a first memory unit associated with said first input unit to a corresponding value from said first set of data;
• one or more layers of stochastic units, comprising a first stochastic unit;
• wherein said first stochastic unit has one or more links traceable to a second subset of said multiple input units, wherein a local coverage extent associated with said first stochastic unit limits said one or more links;
• wherein said one or more links represent first weighted connections between said first stochastic unit and second units;
• wherein said second units are among said second subset of said multiple input units or from said one or more layers of stochastic units between said first stochastic unit and said multiple input units;
• wherein said one or more processors set a memory unit associated with said first stochastic unit to a first value generated based on weighted sum of values of memory units associated with said second units per said first weighted connections represented by said one or more links;
• wherein said one or more processors detect sub-features associated with a higher layer of said one or more layers of stochastic units, based on sub-features detected by a lower layer of said one or more layers of stochastic units;
• one or more output units;
• wherein said one or more processors set a memory unit associated with said one or more output units to a value indicating a detected feature based on one or more said detected sub-features associated with one or more layers of one or more units of said one or more layers of stochastic units;
• wherein said one or more processors determine an energy measure corresponding to said first set of data, based on factors comprising of said first weighted connections, said one or more layers of stochastic units, and said first set of data clamped to said first subset of said multiple input units;
• wherein in said determining said energy measure, a contribution of said energy measure from said first stochastic unit is reduced by a coverage factor of said first stochastic unit;
• wherein said coverage factor of said first stochastic unit is determined based on a value representing fraction of input units within said second subset of said multiple input units traceable from said first stochastic unit that are also in said first subset of said multiple input units clamped to said first set of data;
• wherein said one or more processors output to a second memory unit a value representing a quality measure associated with said detected feature determined based on said energy measure and a baseline;
• wherein said local coverage extent associated with said first stochastic unit is defined by a patch shape covering said second units.