Process for automatic image recognition

• US 4,545,067 A
• Filed: 01/31/1984
• Issued: 10/01/1985
• Est. Priority Date: 01/31/1983
• Status: Expired due to Fees

• Current Assignee: Commissariat a L'Energie Atomique (French Alternative Energies and Atomic Energy Commission)
• Original Assignee: Commissariat a L'Energie Atomique (French Alternative Energies and Atomic Energy Commission)
• Inventors: Juvin, Didier, Dupeyrat, Benoit
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Mancuso, Joseph
• Application Number:
  US06/575,661
• Time in Patent Office: 609 Days
• Field of Search: 382/21, 382/22, 382/56, 358/260, 358/263
• US Class Current: 382/153
• CPC Class Codes: G06V 10/20 Image preprocessing
• 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
• CPC Class Codes: G06V 10/75 Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries

Abstract

A process for the automatic recognition of an image (I) on the basis of a corresponding reference image, including obtaining, by optoelectronic detection means (C), signals representing both the image to be recognized and the reference image, digitizing (in CAN) and recording the respective signals in a memory (M), using a processing unit (UT) to determine the coordinates of respective characteristic contour lines in the image to be recognized and the reference image, segmenting the respective contour lines and coding the segments (Si), attributing to each segment a pair of values (ρ_i, θ_i) relating to its length (ρ_i) and its angle (θ_i) compared with a reference direction, and comparing the respective pairs of values while seeking coincidence between the large segments of the contour to be recognized and the reference contour. The process has application to the recognition of shapes, particularly in robotics.

Claims

1. A process for the automatic recognition of an image (I) on the basis of a corresponding reference image, comprising:

obtaining by optoelectronic detection means for the reference image and the image to be recognized, analog signals whose amplitudes respectively depend on the intensity levels of light rays reflected from the respective images,

digitizing the analog signals into digital values and storing the digital values in a memory,

determining for both the reference image and the image to be recognized, by means of a processing unit (UT) which receives the respective stored digital values, digital values corresponding to the coordinates of the points of at least one contour line (L) characterizing the respective images in a reference plane (XY), the values of these coordinates then being stored,

coding, by means of a reference unit (UT) receiving the digital values of the coordinates of the points of the contour line (L) for both the reference image and the image to be recognized, successive segments (Si), each having ends located on said contour line, said coded digital values being then stored at the same time as the digital values of the coordinates of the ends of the corresponding segments,

obtaining, for each segment of the contour of the image to be recognized and of the contour of the reference image and on the basis of their ends, a pair of characteristic values ρ

i , θ

i, which respectively correspond in curvilinear coordinates to the length ρ

i of each segment and to the angle θ

i formed by said segment with respect to a reference direction, wherein the characteristic values ρ

i, θ

i for the segments (Si) of the reference contour and the contour to be recognized define respective developed images (DL1, DL2),

comparing the pairs of characteristic values obtained respectively for the segments of the contour of the image to be recognized and of the contour of the reference image, wherein the comparing comprises investigating, on the basis of these pairs, coincidences between large corresponding characteristic segments of the reference contour and the contour to be recognized, wherein a large segment is defined relative to a contour as a segment which can be inscribed in a rectangle of limiting dimensions Δ

ρ

_s, Δ

θ

_s, a large segment for which Δ

ρ

_s is larger than Δ

θ

_s being said to be a rectilinear characteristic of the contour, wherein Δ

ρ

_s is locally the greatest length of a segment which is compatible with an angular variation Δ

θ

_s, a large segment for which Δ

θ

_s is larger than Δ

ρ

_s being said to be the angle variation characteristic of the contour, wherein Δ

θ

_s is locally the greatest angular variation compatible with a segment length Δ

ρ

_s, a large segment A of the reference contour coinciding with a large segment B of the contour to be recognized for a displacement (ρ

_o, θ

_o), if by making the origin of the reference large segment A and the point (ρ

_o, θ

_o) in the large segment B coincide, the following properties are proved in curvilinear coordinates

space="preserve" listing-type="equation">ρ

.sub.A -ρ

.sub.B -ρ

.sub.o <

Δ

ρ

.sub.1

space="preserve" listing-type="equation">θ

.sub.A -θ

.sub.B -θ

.sub.o <

Δ

θ

.sub.1

wherein Δ

ρ

₁ and Δ

θ

₁ are fixed.

2. A process according to claim 1, wherein the investigating step comprises:

developed image DL₂ of the contour to be recognized to a length equal to the length of the developed reference image DL₁ of the reference contour by means of a homothetic transformation of its evolute of ratio L₁ /L₂ and then, for these two images centered around the mean value (θ

_o) and both having the same length (L₁), determining the area (A) separating said two images, then comparing said area (A) with a threshold, and then deciding when said area (A) is below said threshold that the image to be recognized corresponds to the reference image.