Nantes Université

Skip to content
Extraits de code Groupes Projets
Vue normale Lien permanent
README.html 25,8 ko
Newer Older
Richard Zanibbi's avatar
Major update Oct 1-11, 2014 (RZ). Modifications summarized in "CHANGES"
Richard Zanibbi a validé
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 
<!doctype html>
<html>
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
<style>
h1,
h2,
h3,
h4,
h5,
h6,
p,
blockquote {
    margin: 0;
    padding: 0;
}
body {
    font-family: "Helvetica Neue", Helvetica, "Hiragino Sans GB", Arial, sans-serif;
    font-size: 13px;
    line-height: 18px;
    color: #737373;
    background-color: white;
    margin: 10px 13px 10px 13px;
}
table {
	margin: 10px 0 15px 0;
	border-collapse: collapse;
}
td,th {	
	border: 1px solid #ddd;
	padding: 3px 10px;
}
th {
	padding: 5px 10px;	
}

a {
    color: #0069d6;
}
a:hover {
    color: #0050a3;
    text-decoration: none;
}
a img {
    border: none;
}
p {
    margin-bottom: 9px;
}
h1,
h2,
h3,
h4,
h5,
h6 {
    color: #404040;
    line-height: 36px;
}
h1 {
    margin-bottom: 18px;
    font-size: 30px;
}
h2 {
    font-size: 24px;
}
h3 {
    font-size: 18px;
}
h4 {
    font-size: 16px;
}
h5 {
    font-size: 14px;
}
h6 {
    font-size: 13px;
}
hr {
    margin: 0 0 19px;
    border: 0;
    border-bottom: 1px solid #ccc;
}
blockquote {
    padding: 13px 13px 21px 15px;
    margin-bottom: 18px;
    font-family:georgia,serif;
    font-style: italic;
}
blockquote:before {
    content:"\201C";
    font-size:40px;
    margin-left:-10px;
    font-family:georgia,serif;
    color:#eee;
}
blockquote p {
    font-size: 14px;
    font-weight: 300;
    line-height: 18px;
    margin-bottom: 0;
    font-style: italic;
}
code, pre {
    font-family: Monaco, Andale Mono, Courier New, monospace;
}
code {
    background-color: #fee9cc;
    color: rgba(0, 0, 0, 0.75);
    padding: 1px 3px;
    font-size: 12px;
    -webkit-border-radius: 3px;
    -moz-border-radius: 3px;
    border-radius: 3px;
}
pre {
    display: block;
    padding: 14px;
    margin: 0 0 18px;
    line-height: 16px;
    font-size: 11px;
    border: 1px solid #d9d9d9;
    white-space: pre-wrap;
    word-wrap: break-word;
}
pre code {
    background-color: #fff;
    color:#737373;
    font-size: 11px;
    padding: 0;
}
sup {
    font-size: 0.83em;
    vertical-align: super;
    line-height: 0;
}
* {
	-webkit-print-color-adjust: exact;
}
@media screen and (min-width: 914px) {
    body {
        width: 854px;
        margin:10px auto;
    }
}
@media print {
	body,code,pre code,h1,h2,h3,h4,h5,h6 {
		color: black;
	}
	table, pre {
		page-break-inside: avoid;
	}
}
</style>
<title>Label Graph Evaluation Tools (LgEval)</title>

</head>
<body>
<!-- MathJax Support for in-line LaTeX -->




<script type="text/x-mathjax-config">
MathJax.Hub.Config({
    tex2jax: {
        inlineMath: [['$','$'], ['\\(','\\)']],
        skipTags: ['script', 'noscript', 'style', 'textarea', 'pre'] // removed 'code' entry
    }
});
MathJax.Hub.Queue(function() {
    var all = MathJax.Hub.getAllJax(), i;
    for(i = 0; i < all.length; i += 1) {
        all[i].SourceElement().parentNode.className += ' has-jax';
    }
});
</script>


<script type="text/javascript" src="http://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML">
</script>


<hr />

<h1>Label Graph Evaluation Tools (LgEval)</h1>

<h2>Version 0.3.2 (October 2014)</h2>

<hr />

<p><em>Copyright (c) 2012-2014, Richard Zanibbi and Harold Mouchère</em></p>

<p><a href="http://www.cs.rit.edu/~rlaz" title="Richard Zanibbi's Home Page">Richard Zanibbi</a> (rlaz@cs.rit.edu)<br/>
Document and Pattern Recognition Lab,
Rochester Institute of Technology, USA</p>

<p><a href="http://www.irccyn.ec-nantes.fr/~mouchere" title="Harold Mouchère's Home Page">Harold Mouchère</a> (harold.mouchere@univ-nantes.fr)<br/>
IRCCyN/IVC Lab, University of Nantes, France</p>

<hr />

<h2>License</h2>

<p><em>These tools are provided 'as is' without any guarantee of suitability for non-research use.  No commercial use is permitted. The tools are being distributed under a Creative Commons license (please see the LICENSE file, and the directory cc_license contains a file detailing the specifics of the license).</em></p>

<h2>Contents</h2>

<ul>
<li><a href="#notes">Notes</a></li>
<li><a href="#purpose">Purpose</a></li>
<li><a href="#install">Installation</a></li>
<li><a href="#lgdescription">Label Graph File Format</a>

<ul>
<li><a href="#primitive">I. Primitive Format</a></li>
<li><a href="#object">II. Object Format</a></li>
<li><a href="#multiple">III. Multiple Levels of Structure</a></li>
</ul>
</li>
<li><a href="#tools">LgEval Tools</a></li>
<li><a href="#other">Additional Programs</a></li>
<li><a href="#refs">References</a></li>
</ul>


<h2><a name="notes"></a>Notes</h2>

<ol>
<li><p>For those wishing to use LgEval with the CROHME competition data, you will also need to install CROHMELib, which is provided separately.</p></li>
<li><p>We also ask that you cite the following paper describing label graphs and associated
 metrics in any publications that you produce using LgEval:</p>

<p> R. Zanibbi, H. Mouchere, and C. Viard-Gaudin (2013) <a href="http://www.cs.rit.edu/~rlaz/files/ZMV_DRR2013.pdf" title="DRR 2013 Paper on Label Graph Metrics">Evaluating Structural Pattern Recognition for Handwritten Math via Primitive Label Graphs</a>. Proc. Document Recognition and Retrieval, Proc. SPIE vol. 8658, pp. 17-1 - 17-11, San Francisco, CA.</p></li>
<li>Information about file differences, metrics and key data structures used for evaluation are provided in the <a href="README_MetricsData.txt" title="README: Metrics and Data">README_MetricsData.txt</a> file.</li>
</ol>


<h2><a name="purpose"></a>Purpose</h2>

<p>The Label Graph Evaluation tools (LgEval) were originally used for scoring handwritten math recognition systems for the <a href="http://www.cs.rit.edu/~rlaz/files/ICDAR_CROHME2013.pdf" title="CROHME 2013 ICDAR paper">CROHME 2013</a> and CROHME 2014 competitions. CROHME is short for <a href="http://www.isical.ac.in/~crohme/index.html" title="CROHME Competition Web Page">Competition on Recognition of Online Handwritten Mathematical Expressions</a> which has been run annually between 2011 and 2014.
For CROHME, the library was used to obtain stroke and symbol-level evaluation of handwritten math expressions. However, label graphs are a very general formalism, and may be used to represent and evaluate structure for many other problems.</p>

<p>A <em>label graph</em> is simply a labeled directed graph. Both nodes and edges are labeled, representing the grouping of input primitives into objects (e.g. grouping strokes into symbols), object types (e.g. symbol names) along with relationships between objects. The section <a href="#lgdescription">Label Graph Files</a> describes the representation in detail. The current version of the library may be used to represent and evaluate multiple levels of structure (e.g. for matrices, which contains symbols, cells, rows, and columns).</p>

<p>Label graphs allow an <em>absolute</em> difference between two structural interpretations to be computed, even when the segmentation of primitives into objects disagree, and when primitives are missing in one or other interpretation. This difference is obtained directly from disagreeing edge and node labels, along with associated Hamming distances (i.e. counting disagreeing node and/or edge labels). Input primitives are assumed to be a fixed set, but can represent any object (e.g. connected components, bounding boxes, pixels in an image, or a combination of these).</p>

<p>In addition to metrics, the library provides visualization of label graphs at the primitive and object levels using the <strong>lg2dot</strong> program. Additional information about label graphs and CROHME may be found in the <a href="#refs">References</a> section.</p>

<h2><a name="install"></a>Installation</h2>

<p><strong>Dependencies:</strong> Before you install LgEval, make sure that the following have also been installed on your system.</p>

<ol>
<li>bash</li>
<li>perl (<strong>with LibXML</strong>)</li>
<li>python 2.6/2.7</li>
<li>Graphviz (for 'dot')</li>
<li>TXL (www.txl.ca)  (required for CROHMELib translation tools)</li>
</ol>


<p>Make sure that <em>CROHMELibDir</em> and <em>LgEvalDir</em> are defined in
   your shell enviroment, e.g. by including the following in your .bashrc  initialization script for bash shell. The last line adds the
   tools to your search path.</p>

<pre><code>export LgEvalDir=&lt;path_to_LgEval&gt;
export CROHMELibDir=&lt;path_to_CROHMELib&gt;     
export PATH=$PATH:$CROHMELibDir/bin:$LgEvalDir/bin
</code></pre>

<p>   To avoid warnings about stack sizes from TXL, issue the command below from the bash command line (<strong>Note:</strong> this requires root user priveleges). This increases the maximum call stack size for the shell. You may also be able to add this to your .bashrc or other shell initialization script.</p>

<pre><code>sudo ulimit -s 16384
</code></pre>

<hr />

<h2><a name="lgdescription"></a>Label Graph File Format</h2>

<p>   There are two formats that may be used to represent a label graph, which may be combined in a single file. Additional example .lg files are provided in the <em>src/Tests/</em> subdirectory.</p>

<h3><a name="primitive"></a>I. Primitive Format (Adjacency Matrix)</h3>

<p>This format introduced for CROHME 2013 is the most basic. The .lg file
defines an adjacency matrix for a labeled graph, where
self-edges are node labels. The file defines nodes with identifiers
and labels along with edge labels, with any unspecified labels being
assigned a default value (underscore, '_').</p>

<p>Nodes that belong to the same object are represented by directed edges
labeled '*' between all pairs of nodes in the object. For example, all
strokes in a symbol are represented by directed '*' edges between all pairs
of strokes (nodes).</p>

<p>Relationships between objects are represented by edges from all nodes in the
parent object of the relationship to every node in the child object of the
relationship.  For example, a 'Right' relationship may be defined using a
labeled directed edge from every node in the object at left to every stroke
in the symbol at right. Undirected relationships are represented by a pair
of directed edges between nodes (e.g. for the '*' relationship defining groupings of nodes into objects).</p>

<p>For CROHME, nodes are used to represent strokes, which are grouped into
symbols (objects), with spatial relationships defined between symbols. It is
assumed that every stroke belongs to exactly one symbol. Here is an example of <em>2+2</em> provided with LgEval (<em>src/Tests/2p2_simple.lg</em>). There are four strokes (primitives),
named <em>s1-s4</em>. The 'right' relationship is represented using the label <em>Right</em>, and the
merging of strokes into a symbol by the label <em>*</em>.</p>

<pre><code># 2 + 2 (Primitive format)
# Four nodes (strokes, with symbol labels)
# FORMAT:
# N, Primitive ID, Label, Weight
N, s1, 2, 1.0
N, s2, +, 1.0
N, s3, +, 1.0
N, s4, 2, 1.0

# Edges
# First, undirected merge edge (two directed edges)
# Strokes s2 and s3 form a '+'
# FORMAT:
# E, Primitive ID (Parent), Primitive ID (Child), Label, Weight
E, s2, s3, *, 1.0
E, s3, s2, *, 1.0

# Finally, four relationship edges for
# 2 -Right-&gt; + -Right-&gt; 2
E, s1, s2, Right, 1.0
E, s1, s3, Right, 1.0
E, s2, s4, Right, 1.0
E, s3, s4, Right, 1.0
</code></pre>

<p>An advantage of this representation is that differences between
interpretations can be defined based on disagreeing labels between two
adjacency matrices, with this difference represented in a third adjacency
matrix. This is useful particularly when the groupings of nodes into objects
differs between interpretations.</p>

<p>The graph represented in the .lg file above is shown below. This image was produced using the <strong>lg2dot</strong> tool. Strokes are shown as nodes, and relationships
between strokes as edges.
 <center>
 <img align=center border=1 height=200 src="doc/readme/2p2_tree.png" alt="Primitive Label Graph for 2+2"></p>

<p> </center><strong>Fig. 1. Primitive Label Graph for '$2+2$' Written with Four Strokes. Strokes are represented by nodes labeled s1-s4.</strong></p>

<h3><a name="object"></a> II. Object Format</h3>

<p>   In this representation, an object and its type are defined by a labeled list of primitive identifiers (e.g. the set of strokes in a
   symbol and the symbol type), along with relationships
   given by labeled edges between objects.</p>

<p>   This is a more compact representation than the 'raw' adjacency matrix
   representation. There is no need to define merge ('*') edges, and edges are defined between objects rather than primitives. Here is our <em>2+2</em> example again, but this time using the object relationship format.</p>

<pre><code># 2 + 2 (Object format)
# 3 objects (symbols)
# FORMAT:
# O, Object ID, Label, Weight, List of Primitive IDs (strokes in a symbol)
O, 2_a, 2, 1.0, s1
O, +_a, +, 1.0, s2, s3
O, 2_b, 2, 1.0, s4

# Relationships (2 object edges)
# FORMAT:
# R, Object ID (Parent), Object ID (Child), Label, Weight   - OR -
# EO, Object ID (Parent), Object ID (Child), Label, Weight
R, 2_a, +_a, Right, 1.0
R, +_a, 2_b, Right, 1.0
</code></pre>

<p>   This format is similar to the one used in the University of Waterloo (Canada)
   <a href="https://www.scg.uwaterloo.ca/mathbrush/corpus.shtml" title="Mathbrush corpus"> MathBrush handwritten math corpus </a> created by Scott MacLean et al.</p>

<h3><a name="multiple"></a> III. Multiple Levels of Structure</h3>

<p>In CROHME 2014, the ability for
nodes to belong to more than one object was added, by allowing a set of
labels to be defined for each node and edge in the underlying adjacency matrix.
For LgEval to handle this, <strong>object types and relationships must be distinct between
levels of structure, and object labels must be distinct from relationship labels.</strong></p>

<p>For
example, it would be a mistake to use the label 'R' for both the symbol 'R'
and the 'Right-of' spatial relationship. Similarly, using 'Right' to
represent the left-to-right order of symbols on a baseline and the
left-to-right ordering of cells can lead to problems, as it it may confuse
which objects at what level of structure are intended for a given relationship.</p>

<p><strong>The merging of nodes into objects is also represented differently.</strong> Each node
and edge between nodes in an object have the same label
(e.g.  for CROHME, all nodes and edges for a handwritten <em>x</em> are labeled <em>x</em>). Provided that labels across structural levels are distinct,
this allows symbols, cells, and rows in a matrix to be distinguished using a
single labeled adjacency matrix.</p>

<p>Below is an example (<em>/src/Tests/MultiLevelExample.lg</em>) illustrating the representation
used to accomodate vectors and matrices for CROHME 2014.
The example is a handwritten vector, a matrix with one row containing x squared and 1.
Rather than two levels of structure as before for primitives and objects, here there are multiple levels of structure arranged in a hierarchy. From bottom to top of the hierarchy, we have:</p>

<ol>
<li>Primitives (six: strokes <em>s1-s6</em>)</li>
<li>Symbols (five: <em>[_1, x_1, 2_1, 1_1, and ]_1</em>) comprised of Primitives;</li>
<li>Cells (two: <em>Cell_1, Cell_2</em>) comprised of Symbols;</li>
<li>Rows (one: <em>Row_1</em>) and Columns (two: <em>Col_1, Col_2</em>) comprised of Cells;</li>
<li>Matrices (one: <em>Vec_1</em>) comprised of Cells;</li>
<li>The top-level expression structure comprised of matrices and symbols.</li>
</ol>


<p>While it is natural to think of matrices as containing cells, cells containing symbols, etc.,  note that the
<strong>representation of objects is defined using strokes
(i.e. input primitives belonging to each object)</strong>. To limit the size of the file, the containment of symbols in cells, cells in rows, cells in matrices etc. is implicit, as this can be recovered from other information.
At the top level of the expression, cells in a matrix are treated as a unit. It is assumed that there are no empty cells in matrices for CROHME 2014.</p>

<pre><code># Example of multiple levels of structure: vector [ x^2 1 ]
# This represents Symbols, Cells, Rows, Columns, Matrix and
# the top-level expression structure.
# Symbols (level above primitives (strokes))
O, [_1, [, 1.0, s1
O, x_1, x, 1.0, s2, s3
O, 2_1, 2, 1.0, s4
O, 1_1, 1, 1.0, s5
O, ]_1, ], 1.0, s6

# Symbol layout (within cells)
R, x_1, 2_1, Sup, 1.0

# Cells (level above symbols)
O, Cell_1, Cell, 1.0, s2, s3, s4
O, Cell_2, Cell, 1.0, s5

# Rows (1) and Columns (2)
O, Row_1, Row, 1.0, s2, s3, s4, s5
O, Col_1, Col, 1.0, s2, s3, s4
O, Col_2, Col, 1.0, s5

# Vector Grid (contains all strokes in cells)
O, Vec_1, Matrix, 1.0, s2, s3, s4, s5

# Layout of Cells in our one row, and all cells
# for both columns.
R, Cell_1, Cell_2, NextCell-Row, 1.0
R, Col_1, Col_2, NextCol, 1.0

# Layout of expression at top level (matrices and symbols 
# outside of matrices)
R, [_1, Vec_1, Right, 1.0
R, Vec_1, ]_1, Right, 1.0
</code></pre>

<p>An illustration of the resulting label graph obtained using <strong>lg2dot</strong> is shown below. Multiple labels for nodes and edges are shown with lists (e.g. "Cell Col Matrix Row Sup" near the top of the graph for an edge from <em>s2</em> to <em>s4</em>). Where two strokes have a directed edges between them with the same label (i.e. a symmetric relationship), a single edge with two arrows is used (e.g. for "Cell Col Matrix Row x" in both directions between <em>s2</em> and <em>s3</em>).</p>

<center>
<img align=center border=1 width=600 src="doc/readme/multilevel.png" alt="Primitive Label Graph for a Vector">  

 </center>


<p><strong>Fig. 2. Primitive Label Graph for '$[x^2 1]$' Written with Six Strokes. Strokes are represented by nodes labeled s1-s6. This graph is complex because of the multiple levels of structure involved. Both edges and nodes may contain multiple labels.</strong></p>

<p>Fig. 2 is very dense, as it represents the grouping of strokes into different objects, such as the strokes inside the vector being identified as belonging to both symbols and cells of the vector. Fig. 3 presents a filtered version of this graph, representing only the vector contents (as an object labeled <em>Matrix</em>), the individual symbols, and the one symbol-level spatial relationship (for $x^2$). As an example of how relationships are represented when there are multiple levels of structure, consider the <em>Right</em> relationships between the brackets and the vector contents. There is a <em>Right</em> edge from the stroke used to draw the left bracket to each stroke in the Matrix object (as seen in Fig. 1). Similarly, the right bracket is represented as being at <em>Right</em> of the Matrix contents by defining an edge from every stroke in the Matrix object to the stroke drawn for the right bracket (again, as seen in Fig. 1).</p>

<center>
<img align=center border=1 width=750 src="doc/readme/multipleSimpler.png" alt="Simpler Primitive Label Graph for a Vector">  

 </center>


<p><strong>Fig. 3. Filtered Version of Fig. 2. The graph shows symbols, the <em>Matrix</em> object defining vector contents, and the superscript relationship for $x^2$.  </strong></p>

<hr />

<h2><a name="tools"></a>LgEval Tools</h2>

<p>The main tools for LgEval are provided in the <em>bin/</em> subdirectory. Call a script without arguments for usage instructions. A doxygen-generated summary of files and classes is available at <a href="doc/html/files.html">doc/html/files.html</a>.</p>

<p><strong>lg2OR and lg2NE</strong><br/>
    Label graph format converters. Each takes a label graph as input, and outputs a label graph in OR or NE format on standard output.</p>

<p><strong>lg2dot</strong><br/>
    Create .dot and .pdf output for a label graph, or visualize the
    difference between two label graphs. Different graph types may be
    produced (requires Graphviz), and differences between
    a pair of graphs may also be visualized. The following graph
    types are supported:</p>

<ul>
<li>Primitive label graphs (e.g. as shown in Figs. 1-3 above)</li>
<li>Bipartite primitive label graphs</li>
<li>Bipartite segmentation graphs for primitives</li>
<li>Directed Acylcic Graph (DAG) represention of objects and object relationships</li>
<li><p>Rooted tree for hierarchical relationship structures (e.g. symbol layout in math expressions)</p>

<p><strong>Note:</strong> The DAG and tree visualizations assume that a single level of structure is
being visualized.</p></li>
</ul>


<p><strong>evaluate, evaluateMat</strong> <br/>
    <em>evaluate</em> is the main evaluation script for label graphs. It
    automatically produces metrics, differences, a result summary,
    and visualizations of recognition errors (requires Graphviz). The program
    produces evaluation results given a directory of output files and
    a corresponding directory of ground truth files, or a user-defined file list.</p>

<p><em>evaluateMat</em> is used to evaluate output for expressions containing matrices
    (used for the matrix recognition task in CROHME 2014).</p>

<pre><code>NOTE: If a node is absent in one of two graphs being compared, it will be inserted as an 'ABSENT' node with unlabeled edges ('_') between the ABSENT node and all other nodes in the graph. See Lg.matchAbsent() in the file lg.py.
</code></pre>

<p><strong>confHist</strong><br/>
Create structure confusion histograms, which show target structures (e.g. of 2-3 symbols) or stroke groups, along with corresponding error graphs and their frequencies. For space, only subgraph errors that are produced more than three times are included in the output. This provides a detailed summary of the specific segmentation and classification errors made by a recognition algorithm. The structure confusion histograms at the object and stroke levels are stored in a (large) .html file.</p>

<p><strong>cdiff, ldiff and vdiff</strong><br/>
Used to compile labeling errors of given types (<em>cdiff</em>), or return the a list of the files containing these errors (<em>ldiff</em>) and view them (<em>vdiff</em>) using 'less.' Regular expression matching over node and edge labels is supported ('egrep' format), and files with or without segmentation errors may be selected for.</p>

<p><strong>lg2lgtree</strong><br/>
    Converts a directory of label graphs using lgfilter (i.e. producing
    trees as output), writing the output files to another directory.</p>

<p><strong>lg2mml</strong><br/>
    Create MathML output from a label graph (requires CROHMELib).</p>

<p><strong>lgfilter</strong><br/>
    Removes non-tree edges from a hierarchically structured label graph (e.g. to obtain symbol
    layout trees from a DAG with inherited relationships for math notation).</p>

<p><strong>crohme2lg</strong><br/>
    Perl program that converts CROHME InkML files (.inkml) to label graph files (.lg).</p>

<p><strong>relabelEdges and relabelOldCROHME</strong><br/>
Tools to replace edge labels in 'old' label graph files using '*' to indicate
merged primitives, but used the same label for symbols and relationships (e.g. using
'R' for both the symbol R and 'Right-of')</p>

<h2><a name="other"></a>Additional Programs</h2>

<p>   All programs below are written in Python (2.x compatible). These
   are located in the <em>src/</em> directory.</p>

<p><strong>metricDist.py</strong><br/>
    Used to select a metric from a CSV file (.m) produced by the 'evallg.py'
    program (used by <strong>evaluate</strong>). Useful for producing histograms.</p>

<p><strong>mergeLg.py</strong><br/>
    Reads two or more .lg files and merges them, printing the result on
    standard output.</p>

<p><strong>lg2txt.py</strong> <br/>
    Convert a graph to a string encoding. Symbol and structure
    mappings are be defined using rules in an accompanying .csv file. An example MathML mapping file is provided in <em>translate/mathMLMap.csv.</em> A (largely incomplete) LaTeX mapping (<em>translate/symbolMap.csv</em>) is also provided.</p>

<p><strong>Unit Tests</strong><br/>
Programs to test functionality for basic label graph operations.</p>

<ul>
<li>testlg.py</li>
<li>testNewSeg.py</li>
</ul>


<hr />

<h2><a name="refs"></a>References</h2>

<p>Further details about label graphs, the metrics used in LgEval and information about the CROHME competitions may be found in the publications listed below. The file README_MetricsAndDataStructures.txt provides some additional information about metrics produced for the raw results, and data structures used in LgEval.</p>

<ul>
<li><p>H. Mouchere, C. Viard-Gaudin, R. Zanibbi and U. Garain. (2014) <a href="http://www.cs.rit.edu/~rlaz/files/Crohme2014FinalVersion.pdf" title="CROHME 2014 ICFHR paper">ICFHR 2014 Competition on Recognition of On-line Handwritten Mathematical Expressions (CROHME 2014)</a>. Proc. Int'l Conf. Frontiers in Handwriting Recognition, Crete, Greece (to appear, Sept. 2014).</p></li>
<li><p>R. Zanibbi, H. Mouchere, and C. Viard-Gaudin (2013) <a href="http://www.cs.rit.edu/~rlaz/files/ZMV_DRR2013.pdf" title="DRR 2013 Paper on Label Graph Metrics">Evaluating Structural Pattern Recognition for Handwritten Math via Primitive Label Graphs Proc. Document Recognition and Retrieval</a>, Proc. SPIE vol. 8658, pp. 17-1 - 17-11, San Francisco, CA.</p></li>
<li><p>H. Mouchere, C. Viard-Gaudin, R. Zanibbi, U. Garain, D.H. Kim and J.H. Kim (2013) <a href="http://www.cs.rit.edu/~rlaz/files/ICDAR_CROHME2013.pdf" title="CROHME 2013 ICDAR paper">ICDAR 2013 CROHME: Third International Competition on Recognition of Online Handwritten Mathematical Expressions</a>. Proc. Int'l Conf. Document Analysis and Recognition, pp. 1428-1432, Washington, DC.</p></li>
<li><p>R. Zanibbi, A. Pillay, H. Mouchere, C. Viard-Gaudin, and D. Blostein. (2011) <a href="http://www.cs.rit.edu/~rlaz/files/MathMetricICDAR2011.pdf" title="ICDAR 2011 Paper on Stroke-Based Performance Metrics for Handwritten Expressions">Stroke-Based Performance Metrics for Handwritten Mathematical Expressions</a>. Proc. Int'l Conf. Document Analysis and Recognition, pp. 334-338, Beijing.</p></li>
</ul>


<hr />

<p><em>This material is based upon work supported by the National Science Foundation (USA) under Grant No. IIS-1016815. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.</em></p>
</body>
</html>