Scale specific and robust line/edge encoding of images

• US 5,684,894 A
• Filed: 11/28/1994
• Issued: 11/04/1997
• Est. Priority Date: 11/28/1994
• Status: Expired due to Term

• Current Assignee: Ip Ot Sub Ulc
• Original Assignee: Eastman Kodak Company
• Inventors: Shustorovich, Alexander
• Primary Examiner(s): Johns, Andrew
• Assistant Examiner(s): Davis, Monica S.
• Application Number:
  US08/345,101
• Time in Patent Office: 1,072 Days
• Field of Search: 382/174, 382/248, 382/199, 382/206, 382/207, 382/291, 382/232, 348/403
• US Class Current: 382/232
• CPC Class Codes: G06T 9/20 Contour coding, e.g. using detection of edges
• CPC Class Codes: G06V 10/46 Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features colour feature extraction G06V10/56

Abstract

Prescribed attributes such as lines and edges are detected in an image wherein the image is digitized to a prescribed spatial and a digital encoding resolution. Directional energy and positions of linear structures in the image are then estimated, and global and local normalization of the linear structures are performed. Spatial and directional competition is conducted among the linear structures and winners of this competition are determined. The winners are encoded into strings of digital data words, with the encoded winners forming a structural description of the original image. The encoded structural description of the original image can be stored and used in other image processing algorithms, such as those used for pattern recognition.

Claims

1. A method of encoding lines and or edges of selectable scale in an image, comprising the steps of:

(a) capturing scale-specific line and or edge direction and position of linear structures of said image independently of resolution of said image; and

(b) encoding said direction and position of said linear structures of said image into an oriented pixel representation of said image;

wherein step (b) comprises encoding said direction and position in said oriented pixel representation of said image by determining projections of said image onto a discrete basis set of two-dimensional filters, wherein said two-dimensional filters comprise wavelets of a type including Gabor wavelets, and wherein said two-dimensional filters model steerable, scaleable and shiftable line and or edge detecting kernels using linear combinations of the detecting kernels.

2. A method according to claim 1, further including the step of:

(c) processing said oriented pixel representation of said image as encoded in step (b) to detect prescribed characteristics of said image.

3. A method of encoding an image comprising the steps of:

(a) digitizing an original image to produce a digitized image;

(b) processing said digitized image to obtain an estimate of direction energy and the position of linear structures in said original image;

(c) performing global and local normalization of the linear structures obtained in step (b);

(d) conducting spatial and directional competition among said linear structures and determining winners of said competition; and

(e) encoding said winners into strings of digital data codes representative of encoded winners, so as to form a structural description of said original image;

wherein step (b) includes the steps of;

(b1) calculating a plurality of projections of said original image onto a local basis set;

(b2) estimating projections onto sine and cosine type wavelets of intermediary sizes and orientations, wherein said sine and cosine type wavelets comprise wavelets of a type including Gabor wavelets;

(b3) determining total energy associated with each wavelet pair;

(b4) estimating a line/edge phase shift; and

(b5) obtaining a value of orthogonal shift relative to a center of a Gaussian envelope from the estimated line/edge phase shift.

4. A method according to claim 3, further including the step of:

(f) processing said structural description formed in step (e) to detect features of said original image.

5. A method according to claim 3, further comprising the step of:

(g) displaying said winners as a function of the strings of data words to form a visualization of the structural description of said original image.

6. A method according to claim 3, wherein step (b) further comprises determining dependence of characteristic size and amplitude of models for wavelets of intermediary sizes and orientation on coefficients of linear combinations of the sine and cosine type wavelets.

7. A method according to claim 3, wherein said original image comprises an alpha-numeric character.

8. A method according to claim 3, wherein said original image comprises a vein pattern.

9. A system for encoding an image comprising:

an image coupling mechanism which is operative to provide said image as a digitized image having a prescribed spatial resolution and a digital encoding resolution;

a digital memory to which said digitized image is coupled for storage therein; and

a digital image processor which is coupled to said digital memory and is operative to process said digitized image in accordance with the following image processing steps;

(a) processing said digitized image to obtain an estimate of direction energy and the position of linear structures in said original image;

(b) performing global and local normalization of the linear structures obtained in step (a);

(c) conducting spatial and directional competition among said linear structures and determining winners of said competition; and

(d) encoding said winners into strings of digital data codes representative of encoded winners, so as to form a structural description of said original image;

wherein step (a) includes the steps of;

(a1) calculating a plurality of projections of said original image onto a local basis set;

(a2) estimating projections onto sine and cosine type wavelets of intermediary sizes and orientations, wherein said sine and cosine type wavelets comprise wavelets of a type including Gabor wavelets;

(a3) determining total energy associated with each wavelet pair;

(a4) estimating a line/edge phase shift; and

(a5) obtaining a value of orthogonal shift relative to a center of a Gaussian envelope from the estimated line/edge phase shift.

10. A system according to claim 9, wherein said digital image processor is operative to process said digitized image in accordance with the following additional image processing step:

(e) processing said structural description formed in step (d) to detect features of said image.

11. A system according to claim 9, further including a display device, and wherein said digital image processor is operative to cause said winners to be displayed by said display device as a function of strings of data words to provide a visualization of the structural description of said original image.

12. A system according to claim 9, wherein, in said digital image processor, step (b) further comprises determining dependence of characteristic size and amplitude of models for wavelets of intermediary sizes and orientation on coefficients of linear combinations of the sine and cosine type wavelets.

13. A system according to claim 9, wherein said original image comprises an alpha-numeric character.

14. A system according to claim 9, wherein said original image comprises a vein pattern.