Automated method and device capable of providing dynamic perceptive invariance of a space-time event with a view to extracting unified semantic representations therefrom
Automated method and device suitable for ensuring the dynamic perceptual invariance of an event with a view to extracting therefrom unified semantic representations are provided. The event is perceived by a linguistic data translator that delivers a signal (HD) referenced (x,y), which signal is transformed into a signal (MAP 1 ) referenced (i,j) through a unit (Dec) that carries out a Gaussian filtering operation that is parameterized by w and decimated by a coefficient k, and transformed into a signal (MAP 2 ) referenced (X,Y) representative of the invariant event through a unit (ROI). A unit ( 71 ′) transforms the signal (MAP 1 ) into a signal of oriented edge (ImBo) and of curvature (ImCb), which are presented to a dynamic attractor ( 80 _ 0 ) that statistically transforms this information into a curvature cb 0 and average orientation bo0, centroid i0,j0 and size of the point cloud ap0,bp0 in order to supply an invariance-computing unit (Inv), which delivers the parameters w, k and the addresses (X,Y) of the signal (MAP 2 ). The invention is applicable in the neuroscience field as an electronically integratable memory-storage unifier.
1. A signal-processing circuit for extracting from a signal a semantic representation, said circuit being configured to carry out:
an iterative computation of a factor k of change of spatial scale and of decimation of said signal, said iterative computation being parameterized in order to keep constant a median curvature value of an oriented edge between two limits chosen in a region of interest selected in a window of said signal;
an application of change-of-frame operations to the coordinates of said region of interest in order to transform said coordinates into a coordinate system of elementary semantic representations;
said processing circuit furthermore being configured to apply to the factor k a homothetic ratio determining a spatial-temporal window depending on a number z of most significant bits of the signal in the region of interest during a period T, said spatial-temporal window being suitable for carrying out comparisons with elementary semantic representations stored in a memory connected to the processing circuit.
2. The signal-processing circuit as claimed in claim 1 , wherein the iterative computation of the factor k comprises an elementary extraction of oriented edges and of curvatures of the signal transformed to the scale 1/k.
3. The signal-processing circuit as claimed in claim 2 , wherein the iterative computation of the factor k furthermore comprises a computation of a looped bilinear spatial-temporal histogram delivering as output, for the region of interest, a median oriented-edge value, a value of the curvature of said oriented edge, a centroid value of a point cloud defined by the points of the decimated signal in the region of interest, and values of the dimensions of the limits of said cloud.
4. The signal-processing circuit as claimed in claim 3 , wherein the change-of-frame operations comprise a geometric origin translation to the centroid of the point cloud after change of scale.
5. The signal-processing circuit as claimed in claim 4 , wherein the homothetic ratio is equal to a power of 1/2, said power being equal to the number of least significant bits between the limits of the cloud of the scaled signal in the coordinate system of elementary semantic representations.
6. The signal-processing circuit as claimed in claim 1 , wherein the change-of-frame operations comprise a rotation by an angle equal to that of the oriented edge with an origin axis of the coordinate system of the elementary semantic representations.
7. The signal-processing circuit as claimed in claim 1 , furthermore configured to compute the factor k of change of scale at the output of a Gaussian filtering operation the matrix of which is defined by a parameter w, said parameter being related to the factor k by an affine relationship.
8. The signal-processing circuit as claimed in claim 7 , wherein the affine relationship is such that k=0.81w+1.09.
9. A device for extracting perceptual invariants from a signal, comprising one or more:
linguistic translators of said signal;
signal-processing circuits as claimed in claim 1 ;
dynamic attractors configured to determine elementary semantic signal representations;
associative memories that are associative via address or content, for storing/using said elementary semantic signal representations.
10. A signal-processing method for extracting from a signal a semantic representation, said method comprising:
an iterative computation of a factor k of change of spatial scale and of decimation of said signal, said iterative computation being parameterized in order to keep constant a median curvature value of an oriented edge between two limits chosen in a region of interest selected in a window of said signal;
an application of change-of-frame operations to the coordinates of said region of interest in order to transform said coordinates into a coordinate system of elementary semantic representations;
an application to the factor k of a homothetic ratio determining a spatial-temporal window depending on a number z of most significant bits of the signal in the region of interest during a period T, said spatial-temporal window being suitable for carrying out comparisons with elementary semantic representations.
11. A method for extracting perceptual invariants from a signal, comprising one or more:
acquisitions of samples of said signal;
applications of a signal-processing method as claimed in claim 10 ;
executions of dynamic-attractor processes in order to determine elementary semantic signal representations;
accesses to one or more associative memories that are associative via address or via content, for storing/using said elementary semantic signal representations.