Neural network for detection and correction of local boundary misalignments between images

• US 5,351,311 A
• Filed: 07/28/1992
• Issued: 09/27/1994
• Est. Priority Date: 07/28/1992
• Status: Expired due to Fees

• Current Assignee: the united states of america as represented by the secretary of the navy
• Original Assignee: the united states of america as represented by the secretary of the navy
• Inventors: Vermillion, Donald, Poston, Wendy L., Rogers, George W., Priebe, Carey E., Solka, Jeffrey L.
• Primary Examiner(s): Chin, Stephen
• Assistant Examiner(s): MAY, TIMOTHY
• Application Number:
  US07/920,700
• Time in Patent Office: 791 Days
• Field of Search: 382/14, 382/15, 382/42, 382/45, 382/44, 382/48, 382/49, 395/21, 358/105
• US Class Current: 382/156
• CPC Class Codes: G06V 10/754 involving a deformation of the sample pattern or of the reference pattern; Elastic matching

Abstract

A network is provided for the detection and correction of local boundary misalignments in a two-dimensional pixel space between a reference and transformed image. An input layer has input layer sections, each of which contains a plurality of input nodes associated with a cell. The cell is centered on a pixel and divided along a straightline orientation into first and second cell sections. Each of the input nodes outputs a digital signal indicative of the presence or absence of a contrast gradient as measured by the two cell sections. A second layer has a plurality of second layer sections, each of which is associated with one of the input layer sections and contains a plurality of second layer nodes. Each second layer node is responsive to a combination of input nodes to indicate the presence or absence of a boundary misalignment between the reference and transformed images. Presence of a contrast gradient at the combination of nodes defines a local boundary misalignment. A third layer has a plurality of third layer nodes, each of which is associated with one of the second layer sections. Each third layer node weights and combines outputs of the second layer nodes to output a signal defining a direction to shift the transformed image perpendicular to the straightline orientation. The third layer are outputs a signal defining the local correction of the local boundary misalignment between the reference and transformed images for the centered pixel.

Claims

1. A neural network for correction of boundary misalignments between a monochrome reference image and a monochrome transformed image that is related to the reference image by an unknown transformation, the reference and transformed images further being defined within a given pixel space wherein, for each pixel in the given pixel space, the neural network comprises:

an input layer having a plurality of input laver sections, each of the input layer section containing a plurality of input nodes encompassing at least one pixel within the given pixel space divided along a predetermined orientation into first and second sections of a cell;

each of said input nodes having, processor means for detecting a unique contrast gradient defined by a digital state comparison between the pixels of the cell sections selected from a group consisting of;

i) the first and second sections viewed with respect to only the reference image, ii) the first and second sections viewed with respect to only the transformed image, iii) the first sections viewed with respect to the reference and transformed images, and iv) the second sections viewed with respect to the reference and transformed images, wherein each of said input nodes outputs a first signal defining one of a presence or absence of the detected contrast gradient as measured by the selected cell sections;

a second layer having a plurality of second layer sections respective associated with one of said input layer sections and containing a plurality of second layer nodes respectively responsive to the outputs of a predetermined combination of said input nodes to output a second signal defining a local boundary misalignment between the reference and transformed images when said first signal from each of said input nodes defines the presence of the measured contrast gradient;

a third layer having a plurality of third layer nodes respectively associated with one of said second layer sections and responsive to the outputs thereof for weighting and combining the outputs of said associated second layer nodes to output a local correctional signal defining a direction to shift the transformed image perpendicular to the predetermined orientation associated with said one of said input layer sections; and

means weighting and averaging the local correctional signal outputs of said third layer for calculating a fourth signal defining global misalignment between the reference and transformed images of the given pixel space.

2. A neural network as in claim 1, wherein each of said second layer nodes comprises a logical AND circuit.

3. A neural network as in claim 1, wherein each of said third layer nodes comprises a summing circuit.

4. A neural network as in claim 1, wherein said means for weighting and averaging the local correctional signal outputs comprises a resistive grid network.

5. The neural network as defined in claim 1 wherein each of the input layer sections is centered on said one of the pixels along one of the predetermined orientations as a straightline.

6. A neural network as in claim 5, wherein the predetermined straightline orientations have relative orientations of 0°

, 45°

, 90° and

135°

, respectively.

7. A neural network as in claim 5, wherein said processor means is sensitive to the unique contrast gradients said comprises four homogeneous processors and four heterogeneous processors.

8. The neural network as defined in claim 1 further including means connected in parallel to the third layer nodes for storing the local correctional signal outputs respectively corresponding to the predetermined orientations of the input layer sections and supplying the same to the weighting and averaging means.

9. In a system for correction of localized and global boundary misalignments between reference and transformed images the reference and the transformed images further being defined within a given pixel space;

a neural network having an input means for detecting contrast gradients corresponding to said reference and said transferred image;

a processing means connected to the input means for determining the localized boundary misalignment from said contrast gradients; and

means connected to the processing means of the neural network for calculating the global boundary misalignments from the determined localized boundary misalignments;

the input means including a plurality of cells, each of said cells being divided into first and second cell sections, the contrast gradients being defined by comparison between the pixels in pairs of the cell sections selected from a group consisting of;

(i) the first and second sections of the cells viewed with respect to only the reference images thereof, (ii) the first and second sections of the cells viewed with respect to only the transformed images thereof, (iii) the first section of the cells viewed with respect to the reference and transformed images thereof, and (IV) the second sections of the cells viewed with respect to the reference and transformed images thereof.