Image processing

French version

Papers and thesis about image processing research. Conception of 2D-symbolic codes reader by image analysis.



PhD thesis: overview

Presentation

Title: Étude sur la lecture automatique de codes bi-dimensionnels par traitement d'image (A study on automatic reading of bi-dimensional symbolic codes by image processing).

Specialization: Image processing (Traitement du signal)

Subject: conception of 2D-symbolic codes reader by image analysis.

Keywords: symbolic codes, edge search, image processing, ligne search, texture segmentation,

Oral presentation did occur on 17 December 1997 at ENSEEIHT.

Summary

Data Matrix symbol Data Matrix is a new kind of bi-dimensional symbolic code issued from the classic bar codes symbols. It uses the 2 directions of the plane, like a chess pad where each cell represents one binary digit. The point of the study is to design reading algorithms for these codes by image processing.

The main constraints are:

Data Matrix image

Many papers have been reviewed, aiming to find and understand the existing methods for these kind of problem.

Comparison of the methods led us to conclude on a exhaustive and coherent method based on edge detection by laplacian detectors, edge localisation by gradient operator, detection validation by texture analysis.

The developed method is:

The method has been implemented in an image processing software. Many tests have been performed to validate and figure the limits of the method against cell size, contrast, heterogeneous background level, optical and projective distortions, focalisation. A detailed study on time processing of the different steps is presented.
Symbol image Read symbol Symbol content
a7n03g21 smb3 163 81 92 94 75 34 90 205 204 128 231 54 193 247 165 157 220 125 226 133 106 123 105 149 219 21 27 149 171 166 71 118 43 67 246 87 4 8 170 249 73 191 21 185 26 219 0  

Oral presentation report

The oral presentation report is only (available in French...)

Credits

ENSEEIHT This research has been conducted in group Signaux Images & Communications () du Laboratoire d'Électronique de l'ENSEEIHT (LEN7LEN7 ) - INPT.
The thesis has been supervised by Pr. Michel Cattoen, professor at ENSEEIHT and founded by intermtc Intermec - Technologies Corporation (), part of the group Unova.
Jury: Dominique Barba - Rapporteur, Jean Bajon - Examinateur, Maurice Briot - Rapporteur, Philippe Marthon - Examinateur, Michel Cattoën - Directeur de thèse, Jean-Louis Massieu - Examinateur.

SIC - LEN7 - ENSEEIHT
2 rue Charles Camichel
B.P. 7122
31071 Toulouse CEDEX 7 (France)

Tel.: +33 (0) 5 61 58 83 20

Contact : Pr Cattoën (cattoen@len7.enseeiht.fr)
http://www.dunwich.org/baptiste/sic/indexe.html

Download

I do not have any more copies of the final report (132 pages), and there will be no reprints. However, I made it available online. You will have to download separately the text (858 kB in zip-file) and the graphics (8,9 MB). In the text, the images are inserted as hard-coded references to an absolute path; you will have to either change the references, either lay the graphics under a location which path would be identical as the one in the document. The report is in French.

Publication: Edge and line detection in low level analysis

This work is part of a study on 2D-symbolic codes reader by image analysis. You may get the full length version.

Reference:
Baptiste Marcel() and Michel Cattoen, « Edge and line detection in low level analysis », 3rd Workshop on Electronic Control and Measuring System, 2-3 June 1997, p. 89-97.

Abstract:
« Edge detection and linear feature extraction are important and critical components in an image understanding system, as the result of the extraction will be the basis of the high level processes.
In this paper, we have described a new straight line detection algorithm, using traditional derivative-based edge operators, which works without internal thresholding, and thus delay detection decisions in the main processes of the image analysis system rather than in the low-level line detection processes.
The algorithm includes two derivative processing by well-known laplacian and gradient filters to determine edge location and orientation, then extract the lines by three steps which are: edge linking, edge straightening and line correction. The usual smoothing and thinning steps are not processed as they are implicit in the algorithm by the choice of the derivative method.
The algorithm is executed in 5 phases. In phase 1, thin edges are detected on zero crossings of the laplacian filter's response. The size of the laplacian filter determines the bandwidth of the implicit smoothing in the filter process. In phase 2, the orientation of each detected edge is computed by 2 gradient filters. In phase 3, the edges are linked according to the direction of their gradient vector, and the short chains of pixel, i.e. smaller then 3 or 4 pixels, are eliminated. This elimination removes from the response all the noise which had been detected by the first phase and thus eliminate the spurious edges, keeping the low contrast lines. In phase 4, the chains are broken to obtain straight lines. In phase 5, lines are corrected to recover isolated missing edge element and to join lines on corner if necessary.
The algorithm works well for detection and localization of Data-Matrix symbols (kind of 2 dimensionnal bar-codes) in images with bimodal intensity distribution.
Several important parts of this method are based on linear filters. Thes computations which usually occur in low-level layers of the system should be able to process fast on hardware embodiment, although this has not yet be verified. »

Key words:

Publication: Calcul de translation et rotation par la transformée de Fourier

« Estimation of Translation and Rotation by Fourier Transform » . This work is the result of a six mounths work on LAAS (CNRS) about image translation and rotation, and is part of a robotic project for autonomous robots. Autonomous robots are up to date, with success of Mars Pathfinder mission.

Reference:
Baptiste Marcel (), Maurice Briot and Rafael Murrieta, « Calcul de translation et rotation par la transformée de Fourier », Traitement du signal (), Vol. 14, n°2, mars 1997, p. 135-149.

Abstract:
« In the research area of vision-aided motion sensors, the rotation parameters can be computed from the motion in the picture. The properties of translation and rotation in the frequency domain of the Fourier transform are used here. This study is restricted to rigid-body transformations, but other application domains, such as matching of rigidly misaligned images, also exist. »

Key words:


Code

I do not believe this code is still usable in an industrial context because it's quite old. However...
For those who whish to have a look, be informed that the application had a temporary work name: Iris. It started as a C application under M$-DOS, then in the middle, we migrated to C++ under M$*Windows.
We had a framework which was able to acquire snapshots from a camera, then the framework was calling functions to process whatever.
In the soup that I leave online, you will have to thread your way around.


Abreviations and links


Scholar background

Engineer in computer science and doctor in electronics.

I graduated engineer at INSA (), in Toulouse, and I made my first research study (D.E.A.) at LAAS - CNRS (Toulouse).

I made my PhD thesis at ENSEEIHT (École Nationale Supérieure d'Électronique, Électrotechnique et Hydraulique de Toulouse), within the group S.I.C. () (Signaux, Images et Communication), part of LEN7 () (Laboratoire d'Électronique de l'ENSEEIHT), in Toulouse (France), in the field of image analysis.


File created July 1996, last update 21/02/2011 by by Baptiste MARCEL (see page Contact), located in Asnières-sur-Seine (France).
If you enjoyed this page, please do not forget to visit my homepage and to request more information about this site.