Conference PaperPDF Available

GREC 2007 Arc Segmentation Contest: Evaluation of Four Participating Algorithms

Authors:

Abstract and Figures

Automatic conversion of line drawings from paper to electronic form requires the recognition of geometric primitives like lines, arcs, circles etc. in scanned documents. Many algorithms have been proposed over the years to extract lines and arcs from document images. To compare different state-of-the-art systems, an arc segmentation contest was held in the seventh IAPR International Workshop on Graphics Recognition - GREC 2007. Four methods participated in the contest, three of which were commercial systems and one was a research algorithm. This paper presents the results of the contest by giving an overview of the dataset used in the contest, evaluation methodology, participating methods and the segmentation accuracy achieved by the participating methods.
Content may be subject to copyright.
GREC 2007 Arc Segmentation Contest:
Evaluation of Four Participating Algorithms
Faisal Shafait1, Daniel Keysers1, and Thomas M. Breuel2
1Image Understanding and Pattern Recognition (IUPR) research group
German Research Center for Artificial Intelligence (DFKI) GmbH
D-67663 Kaiserslautern, Germany
faisal@iupr.dfki.de, keysers@iupr.dfki.de
2Department of Computer Science, Technical University of Kaiserslautern
D-67663 Kaiserslautern, Germany
tmb@informatik.uni-kl.de
Abstract. Automatic conversion of line drawings from paper to elec-
tronic form requires the recognition of geometric primitives like lines,
arcs, circles etc. in scanned documents. Many algorithms have been pro-
posed over the years to extract lines and arcs from document images.
To compare different state-of-the-art systems, an arc segmentation con-
test was held in the seventh IAPR International Workshop on Graph-
ics Recognition - GREC 2007. Four methods participated in the con-
test, three of which were commercial systems and one was a research
algorithm. This paper presents the results of the contest by giving an
overview of the dataset used in the contest, evaluation methodology,
participating methods and the segmentation accuracy achieved by the
participating methods.
Keywords: Graphics Recognition, Line Drawings, Technical Drawings,
Arc Segmentation Contest.
1 Introduction
Reliable detection of geometric primitives like lines, arcs, circles etc. in document
images is one of the key problems in graphics recognition. Due to the importance
of this task, the International Association for Pattern Recognition’s Technical
Committee on Graphics Recognition (IAPR TC10) has been organizing biennial
arc segmentation contests since 2001 [1,2,3]. The purpose of these contests was
to provide a platform for comparative evaluation of state-of-the-art research and
commercial graphics recognition algorithms. The benchmarking of algorithms in
this way helps in objectively evaluating the performance of participating sys-
tems and highlights the strengths and weaknesses of these systems. Therefore
the contest-based approach for comparing algorithms is also used in other do-
mains of document analysis research, like page segmentation [4], handwriting
recognition [5], and document image dewarping [6].
This contest is fourth in the series of arc segmentation contests and was held
at the seventh IAPR International Workshop on Graphics Recognition (GREC
2007), in Curitiba, Brazil, September 20-21, 2007. A dataset of five training and
five test images was used in the contest. This contest was different from the
previous three contests from the view point of ground-truth representation and
performance evaluation protocol. Previous arc segmentation contests used the
VEC format for representing arcs, and used the VRI score [7] as a measure of
performance of arc segmentation. In this contest, we have used a color-based
representation and evaluation scheme [8] discussed in Section 2 and Section 3.
Four methods participated in the contest. A brief description of the methods is
given in Section 4. The dataset used in the contest and the results of the contest
are given in Section 5 followed by a conclusion in Section 6.
2 Representation of Geometric Primitives
The traditional way of representing geometric primitives like lines, arcs, or circles
in a drawing is to use their parametric representation. The VEC format uses this
representation in plain text form, using one line of parameters for each arc. Other
drawing tools can then read this format and reproduce an image containing
exactly the arcs given in the VEC-format text file. However, if some of the arcs
are incorrect, it is hard to find the source of error, since the correspondence of the
arcs in VEC-format to pixels in the original image can not be easily established.
In addition, the performance of the algorithm can not be judged by looking
only at the VEC-format text file, and specialized software is needed to view the
detected arcs and analyze the segmentation errors.
To overcome these problems, we propose a new representation of arc seg-
mentation. This representation is based on pixel-accurate color-coding of page
segmentation as proposed in [8]. Arc segments in an image are represented within
the image such that each pixel belonging to an arc is assigned as its value the
index of the arc. A particular color can be assigned to the page background
(e.g. 0xffffff) and to all pixels not belonging to any arc (e.g. 0x000000). This
representation of arc segmentation is particularly convenient because it can be
used to accurately represent different arcs in the same image as shown in Fig-
ure 1. Secondly, it can be saved and exchanged using any lossless color image
format, thereby avoiding the need for specialized software for viewing the arcs.
The assignment of colors to arcs is arbitrary, so any colors can be chosen for rep-
resenting different arcs. Pixels belonging to more than one arc can be assigned
a unique color if needed.
3 Vectorial Score for Performance Evaluation
To evaluate the performance of an arc segmentation algorithm, we use the vecto-
rial score proposed in [8]. According to this vectorial score, different errors that
are measured are:
Oversegmented arcs: the number of arcs that are either split into more than
one arc, or are partially detected.
Undersegmented arcs: the number of arcs merged with some other arc.
Fig. 1. An example image to demonstrate color encoding of arc segments. Each
arc found in the image is labeled with a unique color.
Total Oversegmentations: the total number of segmentations that ground-
truth arcs were split into.
Total Undersegmentations: the total number of segmentations that would
be needed to split all merged arcs.
Missed arcs: the number of arcs that were not found by the algorithm.
False alarms: the number of detecting arcs originating from noise or non-
graphics elements.
4 Participating Methods
Results of four methods were presented for participation in the contest:
1. Liu Wenyin’s method [9]
2. Vectory software ver. 5.0 (http://www.graphikon.de)
3. Scan2CAD software ver. 7.5d (http://www.softcover.com)
4. VPstudio software ver. 8.02 (http://www.softelec.com)
Liu Wenyin from City University of Hong Kong provided the results of his
method, whereas Hasan Al-Khaffaf from Universiti Sains Malaysia presented the
results of the other three commercial systems.
5 Results
The results of all participating methods on each test image are shown in Figures 2
to 6. The test images were obtained by scanning selected engineering drawings
from different books. The ground-truth was then generated manually by coloring
all the pixels belonging to an arc with a unique color using an off-the-shelf image
manipulation program.
(a) Original Image (b) Ground-Truth Image
(c) Wenyin’s Result (d) Scan2CAD’s Result
(e) Vectory’s Result (f) VPstudio’s Result
Fig. 2. Results of the four participating methods on the first test image. For
clarity in the ground-truth image, only those foreground pixels that belong to
an arc are shown. Due to the presence of the background grid, none of the
participating methods could correctly segment all arcs from the image.
(a) Original Image
(b) Ground-Truth Image
(c) Wenyin’s Result
(d) Scan2CAD’s Result
(e) Vectory’s Result
(f) VPstudio’s Result
Fig. 3. Results of the four participating methods on the second test image.
All methods correctly found the circles, but also produced many false alarms
originating from the text in the annotations, except the Vectory software which
seems to have removed the text parts prior to arc segmentation.
(a) Original Image (b) Ground-Truth Image
(c) Wenyin’s Result (d) Scan2CAD’s Result
(e) Vectory’s Result (f) VPstudio’s Result
Fig. 4. Results of the four participating methods on the third test image. VP-
studio software had the best results in this case, since it was the only method
that correctly segmented the two concave curves.
(a) Original Image (b) Ground-Truth Image
(c) Wenyin’s Result (d) Scan2CAD’s Result
(e) Vectory’s Result (f) VPstudio’s Result
Fig. 5. Results of the four participating methods on the fourth test image. Liu
Wenyin’s method had the best results for this image since it was the only method
that correctly found the curved corners in the image.
(a) Original Image (b) Ground-Truth Image
(c) Wenyin’s Result (d) Scan2CAD’s Result
(e) Vectory’s Result (f) VPstudio’s Result
Fig. 6. Results of the four participating methods on the fifth test image. Both
VPstudio and Liu Wenyin’s method had comparable results in this case that
were better than those of Vectory and Scan2CAD.
(a) Original segmented image
(b) Segmented image after setting arc width to 10 pixels
Fig. 7. (a) The segmentation result of VPstudio on a test image. Despite the
algorithm working very well in segmenting the arcs, the evaluation result re-
ported that all 13 arcs were oversegmented in this image, since the results were
supplied with a constant line width of one pixel. (b) The segmentation result
after setting arc width to 10 pixels. The evaluation result for this image reported
no segmentation errors.
Table 1 shows the vectorial score obtained by all participating methods on
the test images. The results were obtained by using a relative threshold of 0.1
and an absolute threshold of 100 pixels. This implies that a segmentation error
was considered significant only if the number of in-correctly segmented pixels
was either larger than 10% of the pixels belonging to an arc or was larger than
100 pixels in total. The results show that all the algorithms over-segmented the
arcs. This happened when all pixels belonging to an arc were not assigned to the
arc by the algorithm. One major reason for this was that the VEC-files for the
commercial systems were supplied with a constant line width of one pixel for all
the arcs. An example showing the segmentation results of the VPstudio software
on a test image is shown in Figure 7(a). Evaluation result for this image reported
that all 13 arcs were over-segmented. To see the influence of this problem, we
re-ran the evaluation using a constant line width of 10 pixels for all systems.
Since we use only the foreground pixels while ignoring the background pixels,
setting the arc width to 10 pixels has the effect of actually ignoring the arc
width. The effect of setting the arc width of all arcs to 10 pixels for the example
image of Figure 7(a) is shown in Figure 7(b). It can be seen that all the pixels
belonging to an arc are now correctly assigned to that arc. Table 2 shows the
evaluation results by ignoring the arc width. This table shows that most of the
over-segmentations were due to the small line thickness supplied by the systems.
Table 1. Different types of errors made by each algorithm on the test im-
ages. The column labels are: total oversegmentations (To), total undersegmenta-
tions (Tu), oversegmented components (Co), undersegmented components (Cu),
missed components (Cm), false alarms (Cf)
Algorithm ToTuCoCuCmCf
Wenyin’s method 21 8 13 6 1 93
Scan2CAD 72 9 48 7 9 64
Vectory 54 9 43 9 14 0
VPstudio 55 4 49 3 8 64
Table 2. Different types of errors made by each algorithm on the test images
when arc width was set to a constant value of 10 pixels in the output of all
algorithms.
Algorithm ToTuCoCuCmCf
Wenyin’s method 17 9 9 6 1 94
Scan2CAD 36 10 20 6 9 66
Vectory 35 13 26 9 13 1
VPstudio 7 5 7 4 8 62
The results show that Wenyin’s method and VPstudio software worked very
well in segmenting arcs from the images and did uniformly better than the other
two systems on most of the performance measures. The number of false alarms
were high for these systems because they did not remove text parts in the images
prior to arc recognition. From that aspect Vectory software performed the best
by removing all textual components from the image, thereby resulting in no
false alarms. Interestingly, for the test image shown in Figure 3, evaluation of
the output of Vectory software reported all 13 arcs as over-segmented. A closer
look revealed that the Vectory software also did a skew correction of the image,
thereby slightly moving all circles from their original position. This resulted in all
circles reported as over-segmented. Liu Wenyin’s method had the least number
of missed errors. Inspection of the results revealed that most of the missed error
in commercial systems originated from ground-truth arcs consisting of round
corners as in Figure 5.
6 Conclusion
This paper presented a summary of the GREC 2007 arc segmentation contest.
We described the pixel-accurate color-based representation of arc segmentation
that was used in the competition along with a vectorial score for measuring
arc segmentation accuracy. The vectorial score enables us to evaluate different
aspects of an arc segmentation algorithm. One research algorithm by Liu Wenyin
and three commercial systems namely Scan2CAD, Vectory, and VPstudio were
presented for participation in the contest. Results showed that Wenyin’s method
and VPstudio out-performed the other two systems, whereas the performance of
Wenyin’s method and that of VPstudio software was not significantly different
from each other.
Acknowledgments
This work was partially funded by the BMBF (German Federal Ministry of
Education and Research), project IPeT (01 IW D03).
References
1. Wenyin, L.: The third report of the arc segmentation contest. In: Proc. 6th IAPR
Workshop on Graphics Recognition, Springer LNCS 3926, Barcelona, Spain (2005)
358–361
2. Wenyin, L.: Report of the arc segmentation contest. In: Proc. 5th IAPR Workshop
on Graphics Recognition, Springer LNCS 3088, Barcelona, Spain (2003) 364–367
3. Wenyin, L., Zhai, J., Dori, D.: Extended summary of the arc segmentation contest.
In: Proc. 4th IAPR Workshop on Graphics Recognition, Springer LNCS 2390. (2001)
343–349
4. Antonacopoulos, A., Gatos, B., Bridson, D.: ICDAR 2007 page segmentation com-
petition. In: Proc. 9th Intl. Conf. on Document Analysis and Recognition, Curitiba,
Brazil (2007) 1279–1283
5. Maergner, V., Abed, H.E.: ICDAR 2007 - arabic handwriting recognition compe-
tition. In: Proc. 9th Intl. Conf. on Document Analysis and Recognition, Curitiba,
Brazil (2007) 1274–1278
6. Shafait, F., Breuel, T.M.: Document image dewarping contest. In: 2nd Int. Work-
shop on Camera-Based Document Analysis and Recognition, Curitiba, Brazil (2007)
181–188
7. Wenyin, L., Dori, D.: A protocol for performance evaluation of line detection algo-
rithms. Machine Vision and Applications: Special Issue on Performance Character-
istics of Vision Algorithms 9(1997) 240–250
8. Shafait, F., Keysers, D., Breuel, T.M.: Pixel-accurate representation and evalu-
ation of page segmentation in document images. In: 18th Int. Conf. on Pattern
Recognition, Hong Kong, China (2006) 872–875
9. Wenyin, L., Dori, D.: Incremental arc segmentation algorithm and its evaluation.
IEEE Trans. on Pattern Analysis and Machine Intelligence 20 (1998) 424–431
... Shafait et al. [15] shifted the way of representing the ground truth data and the performance evaluation. Five real scanned images were used. ...
... 1. There is insufficient research on the effect of noise on raster-to-vector conversion process and the use of an unspecified small amount of noise (such as the study of Wenyin [12]) or performing the test on clean images (such as the study of Shafait et al. [15]). 2. The interaction between noise level and noise removal is not studied in the context of raster-tovector conversion. ...
... This type of image contains straight lines as well as circular arcs. The images from the GREC'03 and GREC'07 contests [11,15] were used because ground truth files were readily available for the performance evaluation task. ...
Article
Many factors, such as noise level in the original image and the noise-removal methods that clean the image prior to performing a vectorization, may play an important role in affecting the line detection of raster-to-vector conversion methods. In this paper, we propose an empirical performance evaluation methodology that is coupled with a robust statistical analysis method to study many factors that may affect the quality of line detection. Three factors are studied: noise level, noise-removal method, and the raster-to-vector conversion method. Eleven mechanical engineering drawings, three salt-and-pepper noise levels, six noise-removal methods, and three commercial vectorization methods were used in the experiment. The Vector Recovery Index (VRI) of the detected vectors was the criterion used for the quality of line detection. A repeated measure ANOVA analyzed the VRI scores. The statistical analysis shows that all the studied factors affected the quality of line detection. It also shows that two-way interactions between the studied factors affected line detection.
... The annotations/texts were removed from the sample images. The ground-truth was marked using pixel-accurate color-coding of page segmentation as presented in [1] and was also used in GREC 2007 Arc Segmentation Contest [2]. The ground-truthed images are shown in Fig. 2These sample images were used for both training and testing. ...
Conference Paper
Recognition of geometric primitives such as line and arc helps in automatic conversion of line drawing document images into electronic form. A large number of raster to vector methods can be found in the literature. A line and arc segmentation contest was held in conjunction with the tenth IAPR International Workshop on Graphics Recognition (GREC 2013) for comparing performance of different methods on a uniform platform. The contest was broken down into two challenges: arc segmentation and line segmentation. The dataset includes engineering drawings (for arc segmentation challenge) and cadastral maps (for line segmentation challenge). Jianping Wu’s method got the highest score (0.541), hence the winner of the Arc Segmentation Contest. Liu Wenyin’s method, the only method participated in the line segmentation contest achieved 66 % segmentation accuracy.
... Only continuous circular arcs were considered in this contest. GREC'07: Shafait et al. [8] evaluated the performance of four vectorization methods: one research prototype (Wenyin's methods) and three commercial software (VPstudio, Scan2CAD, Vectory). Five real scanned images (with no artificial noise) were used in the contest. ...
Full-text available
Article
Empirical performance evaluation of raster to vector methods is an important topic in the area of graphics recognition. By studying automatic vectorization methods we can reveal the maturity of the tested methods whether as a research prototype or a commercial software. Arc Segmentation Contest held in conjunction with the eighth IAPR International Workshop on Graphics Recognition (GREC'09) is an excellent opportu-nity for researchers to present the results of their proposed raster to vector methods. The contest provides a uniform platform where the output of different methods can be analyzed. The relevance of the contest is fur-ther revealed by the creation of new test images with their ground truth data. Hence, researchers with raster to vector methods are encouraged to participate in this contest. This paper presents the plan and suggestions for the intended contest.
... Since [Hilaire and Tombre, 2006] won the 2005 edition of the IAPR International Contest on Circle Detection, [28], building on a comparable approach, had been noticed by the 2007 edition organizers. On their invitation, it competed unofficially (after the contest was closed [Shafait et al., 2008]) and obtained more than decent results (unpublished, since after the deadline, but giving rise to a collaboration with Prof. Liu and and stay at Hong Kong City University in 2007). The final version [12] The most important impact of this work (especially with respect to future research directions) stems from this double participation of the exact same algorithm in two successive editions of the same contest. ...
Full-text available
Thesis
The thesis defended in this document starts with considering some formal approaches to Machine Perception algorithm performance analysis, and how it relates to the limits (and unavoidable subjectivity) of ground truth specification. It establishes the intrinsic ambiguity of interpretation and analysis when used in conjunction with Machine Perception and considers them mainly in relation to Document Image Analysis. After establishing the fact that interpretation is open to ambiguity and that most of this ambiguity comes from inconsistent or different interpretation contexts our overall goal is to investigate whether one can: * establish a form of context description that is appropriate for machine perception (and document image analysis in particular) and whether it can be obtained auto- matically by statistical or formal learning techniques? * use this context description to evaluate algorithm performances? * use this context description to describe data, so that it can be used for information retrieval purposes? * establish formal boundaries or limitations for the previously described descriptions and establish whether there are interpretations that are provably impossible to be obtained through an algorithm. If there is indeed a class of interpretation problems that cannot be solved by an algorithm, the second question would be whether this class can be characterized in some sorts.
... Only solid circular arcs were tested. [9] evaluated the performance of four vectorization methods: one research prototype (Wenyin's method) and three commercial software (VPstudio , Scan2CAD, Vectory). Five real scanned images (with no artificial noise) were used in the contest. ...
Full-text available
Conference Paper
Empirical performance evaluation of raster to vector methods is an important topic in the area of graphics recognition. By studying automatic vectorization methods we can reveal the maturity of the tested methods whether as a research prototype or a commercial software. Arc Segmentation Contest held in conjunction with the eighth IAPR International Workshop on Graphics Recognition (GREC’09) is an excellent opportunity for researchers to present the results of their proposed raster to vector methods. The contest provides a uniform platform where the output of different methods can be analyzed. The relevance of the contest is further revealed by the creation of new test images with their ground truth data. Old documents were used in this contest. Five methods participated (two research prototypes and three commercial software). Two tests were performed namely between-methods test (participated by all methods) and within-method test (participated by only one method). This paper presents the results of the contest. KeywordsPerformance Evaluation-Graphics Recognition-Raster to Vector Conversion Methods-Line Drawings
Chapter
Following the discussion we have made in Chap. 1, in this chapter, graphics recognition: graphical symbol recognition, retrieval and spotting will be discussed. It first provides a clear definition of graphical symbols and tells us where does graphics recognition lie in the DIA. It also extends its discussion to graphics recognition contests (major competitions, organized by international scientific committee in cooperation with the international association for pattern recognition (IAPR)) that have been happening since 90s to see whether the real-world problems have been addressed. Not stopping there, it also quickly explains research standpoints from the author’s point of view.
Chapter
In this chapter a thorough definition of the graph data is provided.
Conference Paper
Circle and circular arc detection in images have been a long standing topic in image analysis. It finds numerous applications for both scanned document images as well as in photographic images. As a result, circle detection algorithms are published regularly and benchmarking data sets and contests have been organized on a regular basis over the last decades. Unfortunately, they have not been able to achieve a very clear image establishing which approaches perform best and under what exact conditions.
Article
In document image processing, arc segmentation plays an important role in vectorization and graphic recognition. Moreover, the unsatisfactory results of several recent arc segmentation contests indicate that conventional methods are inadequate. This paper proposes a new arc segmentation algorithm called SymCAve (an acronym for Symmetry axis, Circle fitting and Average distribution points). First, we locate several seed points and adopt three strategies to ensure that the seed points are proper; then we calculate the center and radius utilizing the seed points. Second, the coordinates of the center and radius are adjusted by employing symmetry axes. Third, the average distribution points method is used to verify whether the points on the circumference are all black pixels. It is a complete circle if all of the points are black pixels. Otherwise, it is a partial circle if some of the points are black pixels and are continuous. Based on this information, the start and end angles of the partial circle can be determined. Finally, these arcs are verified to ensure that the results are accurate. Images and the evaluation tool were obtained from the GREC Workshop's Arc Segmentation contests, to test the systematic performance of the SymCAve algorithm. The experiments demonstrate that the proposed method can provide promising results. However, the algorithm has some drawbacks: it cannot detect a line with width of one pixel, small angles, and any large radius arcs. It is suited for segmenting images with appropriate symmetry axes.
Conference Paper
Empirical performance evaluation of raster to vector conversion is a means of judging the quality of line detection algorithms. Many factors may affect line detection. This paper aims to study scanning resolution of raster images and its effects on the performance of line detection. Test images with three different scanning resolutions (200, 300, and 400 DPI) are vectorised using available raster to vector conversion software. The Vector Recovery Index scores calculated with reference to the ground truth images and the detected vectors are then obtained. These values are analysed statistically in order to study the effects of different scanning resolutions. From the results, Vextractor is found to be better (on average) compared to VPstudio and Scan2CAD. For all the three resolutions, Vextractor and VPstudio perform better than Scan2CAD. Different scanning resolutions affect the software differently. The performance of Vextractor and VPstudio increases from low resolution to moderate resolution, and then decreases with high resolution. The performance of Scan2CAD decreases with the increase in the resolutions.
Full-text available
Article
Dewarping of documents captured with hand-held cam- eras in an uncontrolled environment has triggered a lot of interest in the scientific community over the last few years and many approaches have been proposed. However, there has been no comparative evaluation of different dewarping techniques so far. In an attempt to fill this gap, we have organized a page dewarping contest along with CBDAR 2007. We have created a dataset of 102 documents captured with a hand-held camera and have made it freely available online. We have prepared text-line, text-zone, and ASCII text ground-truth for the documents in this dataset. Three groups participated in the contest with their methods. In this paper we present an overview of the approaches that the participants used, the evaluation measure, and the dataset used in the contest. We report the performance of all par- ticipating methods. The evaluation shows that none of the participating methods was statistically significantly better than any other participating method.
Full-text available
Article
This paper continues the authors' attempt to address the need for objective comparative evaluation of layout analysis methods in realistic circumstances. It describes the Page Segmentation Competition (modus operandi, dataset and evaluation criteria) held in the context of ICDAR2007 and presents the results of the evaluation of three candidate methods. The main objective of the competition was to compare the performance of such methods using scanned documents from commonly- occurring publications. The results indicate that although methods continue to mature, there is still a considerable need to develop robust methods that deal with everyday documents.
Full-text available
Conference Paper
The Arc Segmentation Contest, as the fourth in the series of graphics recognition contests organized by IAPR TC10, was held in association with the GREC'2001 workshop. In this paper we present the extended summary of the contest: the contest rules, performance metrics, test images and their ground truths, and the outcomes.
Full-text available
Conference Paper
This paper presents a new representation and evaluation procedure of page segmentation algorithms and analyzes six widely-used layout analysis algorithms using the proce- dure. The method permits a detailed analysis of the behav- ior of page segmentation algorithms in terms of over- and undersegmentation at different layout levels, as well as de- termination of the geometric accuracy of the segmentation. The representation of document layouts relies on labeling each pixel according to its function in the overall segmen- tation, permitting pixel-accurate representation of layout information of arbitrary layouts and allowing background pixels to be classified as "don't care". Our representations can be encoded easily in standard color image formats like PNG, permitting easy interchange of segmentation results and ground truth.
Full-text available
Conference Paper
This paper describes the Online Arabic handwriting recognition competition held at ICDAR 2009. This first competition uses the ADAB-database with Arabic online handwritten words. This year, 3 groups with 7 systems are participating in the competition. The systems were tested on known data (sets 1 to 3) and on one test dataset which is unknown to all participants (set 4). The systems are compared on the most important characteristic of classification systems,the recognition rate. Additionally, the relative speed of the different systems were compared. A short description of the participating groups, their systems, the experimental setup, and the performed results are presented.
Full-text available
Conference Paper
This paper describes the Arabic handwriting recognition competition held at ICDAR 2007. This second competition (the first was at ICDAR 2005) again uses the IFN/ENIT-database with Arabic handwritten Tunisian town names. Today, more than 54 research groups from universities, research centers, and industry are working with this database worldwide. This year, 8 groups with 14 systems are participating in the competition. The systems were tested on known data and on two datasets which are unknown to the participants. The systems are compared on the most important characteristic, the recognition rate. Additionally, the relative speed of the different systems were compared. A short description of the participating groups, their systems, and the results achieved are finally presented.
Article
This paper describes the Arabic handwriting recognition competition held at ICDAR 2007. This second competition (the first was at ICDAR 2005) again uses the IFN/ENIT- database with Arabic handwritten Tunisian town names. Today, more than 54 research groups from universities, re- search centers, and industry are working with this database worldwide. This year, 8 groups with 14 systems are par- ticipating in the competition. The systems were tested on known data and on two datasets which are unknown to the participants. The systems are compared on the most impor- tant characteristic, the recognition rate. Additionally, the relative speed of the different systems were compared. A short description of the participating groups, their systems, and the results achieved are finally presented.
Conference Paper
The Arc Segmentation Contest, as the sixth in the series of graphics recognition contests organized by IAPR TC10, was held in association with the GREC’2005 workshop. Three systems have participated in the contest. In this paper we present a brief summary: the contest rules, the updated performance metrics, test images and their ground truths, and the outcomes.
Conference Paper
The Arc Segmentation Contest, as the fifth in the series of graphics recognition contests organized by IAPR TC10, was held in association with the GREC'2003 workshop. In this paper we present the report of the contest: the contest rules, performance metrics, test images and their ground truths, and the outcomes.