Multimode invariant processor
First Claim
1. A multimode invariant image processor for classifying patterns in an image, the processor comprising:
- a retina portion for receiving the image and generating in response image gradient information;
a nonlinear processing portion for processing the image gradient information to generate image feature vectors representing image features in the image wherein the nonlinear processing portion comprises a series of neural director layers to aid in discrimination between similar patterns in the image, each layer having at least one neural director, the neural directors in a first of the layers receiving the gradient information from the retina portion and generating in response respective feature vectors each having a dimensionality at least as great as the received gradient information, each successive neural director layer receiving feature vectors generated by neural directors in each previous neural director layer and generating in response respective feature vectors each having a dimensionality at least as great as the received feature vector;
a convergence processing portion for processing the image feature vectors to generate common feature information; and
a classifier portion for receiving the common feature information and generating in response classification information indicating the likelihood that selected features are present in the image.
1 Assignment
0 Petitions
Accused Products
Abstract
A multimode invariant processor is provided to simultaneously classify one or more patterns in multidimensional or in two dimensional “real world” images. The classification is invariant to a translation, a change in scale size and a rotation of a whole or partially hidden photonic image. The multimode invariant image processor comprises a retina portion, a nonlinear processing portion, a convergence processing portion and a classifier portion. The retina portion processes the photonic image to obtain an image data array of pixels and further process the array of pixels through a window difference network to obtain gradients of the image data. The neural directors of the nonlinear processing portion receive the gradients and generate respective feature vectors, which may have a greater dimensionality than the gradient information, to aid in discrimination between similar patterns in the image data. The convergence portion processes the feature information to generate a convergence of common feature information representing at least one image feature in the image data. The classifier portion receives the common feature information and generates in response feature classification information indicating the likelihood that selected features are present in the image.
114 Citations
12 Claims
-
1. A multimode invariant image processor for classifying patterns in an image, the processor comprising:
-
a retina portion for receiving the image and generating in response image gradient information;
a nonlinear processing portion for processing the image gradient information to generate image feature vectors representing image features in the image wherein the nonlinear processing portion comprises a series of neural director layers to aid in discrimination between similar patterns in the image, each layer having at least one neural director, the neural directors in a first of the layers receiving the gradient information from the retina portion and generating in response respective feature vectors each having a dimensionality at least as great as the received gradient information, each successive neural director layer receiving feature vectors generated by neural directors in each previous neural director layer and generating in response respective feature vectors each having a dimensionality at least as great as the received feature vector;
a convergence processing portion for processing the image feature vectors to generate common feature information; and
a classifier portion for receiving the common feature information and generating in response classification information indicating the likelihood that selected features are present in the image. - View Dependent Claims (2, 3, 4, 5, 6)
a positional king-of-the-mountain circuit receiving the feature vectors from the nonlinear processing portion and generating a number of outputs, each output identifying, for one of the feature vectors, a component having the highest value; and
a plurality of summing circuits, each for receiving highest value outputs for like components and generating summed outputs for the like components as the common feature information.
-
-
3. A multimode invariant image processor as defined in claim 2, wherein the convergence processing portion further comprises an interconnection network for receiving like component highest value outputs from the positional king-of-the-mountain circuit and coupling said outputs to a corresponding one of said summing circuits.
-
4. A multimode invariant image processor as defined in claim 2, wherein the summing circuits have a sum threshold value applied thereto such that a summed output for a summing circuit is generated only when the threshold value is exceeded.
-
5. A multimode invariant image processor as defined in claim 1, wherein the classifier portion comprises:
-
a multidimensional memory space partially populated by the common feature information from the convergence processing portion;
a recognition vector arrays having a plurality of recognition vectors in communication with the multidimensional memory space whereby the partially populated multidimensional memory space activates sets of recognition vectors within the recognition array;
a plurality of groups of king-of-the-mountain circuits, each group representing an image primitive of the image in at least one rotational position, each group receiving the corresponding recognition vectors associated with the image primitive and generating at least one group output representing a likelihood that the image contains a primitive in the at least one rotational position;
an angular vector array receiving each group output and generating angular vectors, each angular vector corresponding to a combination of all group outputs of one of the plurality of groups; and
a class multi-king-of-the-mountain circuit receiving the angular vectors and generating the classification information.
-
-
6. A multimode invariant image processor as defined in claim 5, wherein the king-of-the-mountain circuits have a class threshold value applied thereto such that a group output for a king-of-the-mountain circuit is generated only when the class threshold value is exceeded.
-
7. A multimode invariant image processor for classifying patterns in an image, the processor comprising:
-
a retina portion for receiving the image and generating in response image gradient information wherein the retina portion comprises a transducer array receiving the image and transforming the image to an array of pixels, each pixel being represented by a pixel value; and
a window difference network for generating, for each pixel, a gradient vector defining the difference between the pixel value for each pixel and pixel values for selected ones of pixels around each pixel;
a nonlinear processing portion for processing the image gradient information to generate image feature vectors representing image features in the image;
a convergence processing portion for processing the image feature vectors to generate common feature information; and
a classifier portion for receiving the common feature information and generating in response classification information indicating the likelihood that selected features are present in the image. - View Dependent Claims (8, 9, 10)
-
-
11. A multimode invariant image processor for classifying patterns in an image, the processor comprising:
-
a retina portion for receiving the image and generating in response image gradient information;
a nonlinear processing portion for processing the image gradient information to generate image feature vectors representing image features in the image, wherein the nonlinear processing portion comprises a series of neural director layers to aid in discrimination between similar patterns in the image, each layer having a plurality of neural directors, each neural director corresponding to a pixel of the image, each neural director layer corresponding to an image primitive, the neural directors in a first of the neural director layers receiving the gradient information from the retina portion and generating in response respective feature vectors each having a dimensionality at least as great as the received gradient information, each successive neural director layer receiving feature vectors generated by neural directors in each previous neural director layer and generating, in response, respective feature vectors each having a dimensionality at least as great as the received feature vector; and
common feature layers, each corresponding to one of the image primitives, each common feature layer receiving all components of the feature vectors corresponding to the respective image primitive for the common feature layer;
a convergence processing portion for processing the image feature vectors to generate common feature information; and
a classifier portion for receiving the common feature information and generating in response classification information indicating the likelihood that selected features are present in the image. - View Dependent Claims (12)
-
Specification