comparison writeup/techreport.tex @ 477:534d4ecf1bd1

small desription of the font added
author Xavier Glorot <glorotxa@iro.umontreal.ca>
date Sun, 30 May 2010 17:24:26 -0400
parents 5fa1c653620c
children 6593e67381a3
comparison
equal deleted inserted replaced
476:db28764b8252 477:534d4ecf1bd1
427 The performances reported by previous work on that dataset mostly use only the digits. 427 The performances reported by previous work on that dataset mostly use only the digits.
428 Here we use the whole classes both in the training and testing phase. 428 Here we use the whole classes both in the training and testing phase.
429 429
430 430
431 \item {\bf Fonts} 431 \item {\bf Fonts}
432 In order to have a good variety of sources we downloaded an important number of free fonts from: {\tt http://anonymous.url.net}
433 %real adress {\tt http://cg.scs.carleton.ca/~luc/freefonts.html}
434 in addition to Windows 7's, this adds up to a total of $9817$ different fonts that we can choose uniformly.
435 The ttf file is either used as input of the Captcha generator (see next item) or, by producing a corresponding image,
436 directly as input to our models.
437 %Guillaume are there other details I forgot on the font selection?
438
432 \item {\bf Captchas} 439 \item {\bf Captchas}
433 The Captcha data source is an adaptation of the \emph{pycaptcha} library (a python based captcha generator library) for 440 The Captcha data source is an adaptation of the \emph{pycaptcha} library (a python based captcha generator library) for
434 generating characters of the same format as the NIST dataset. The core of this data source is composed with a random character 441 generating characters of the same format as the NIST dataset. The core of this data source is composed with a random character
435 generator and various kinds of tranformations similar to those described in the previous sections. 442 generator and various kinds of tranformations similar to those described in the previous sections.
436 In order to increase the variability of the data generated, different fonts are used for generating the characters. 443 In order to increase the variability of the data generated, different fonts are used for generating the characters.
440 \item {\bf OCR data} 447 \item {\bf OCR data}
441 \end{itemize} 448 \end{itemize}
442 449
443 \subsubsection{Data Sets} 450 \subsubsection{Data Sets}
444 \begin{itemize} 451 \begin{itemize}
445 \item {\bf NIST}
446 \item {\bf P07} 452 \item {\bf P07}
447 The dataset P07 is sampled with our transformation pipeline with a complexity parameter of $0.7$. 453 The dataset P07 is sampled with our transformation pipeline with a complexity parameter of $0.7$.
448 For each new exemple to generate, we choose one source with the following probability: $0.1$ for the fonts, 454 For each new exemple to generate, we choose one source with the following probability: $0.1$ for the fonts,
449 $0.25$ for the captchas, $0.25$ for OCR data and $0.4$ for NIST. We apply all the transformations in their order 455 $0.25$ for the captchas, $0.25$ for OCR data and $0.4$ for NIST. We apply all the transformations in their order
450 and for each of them we sample uniformly a complexity in the range $[0,0.7]$. 456 and for each of them we sample uniformly a complexity in the range $[0,0.7]$.