Mercurial > ift6266

comparison writeup/techreport.tex @ 407:fe2e2964e7a3

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
description des transformations en cours ajout d un fichier special.bib pour des references specifiques
author Xavier Glorot <glorotxa@iro.umontreal.ca>
date Wed, 28 Apr 2010 14:35:01 -0400
parents 4c840798d290
children 6330298791fb
comparison
equal deleted inserted replaced
404:1509b9bba4cc 407:fe2e2964e7a3
70 successful algorithms were proposed to overcome some of these 70 successful algorithms were proposed to overcome some of these
71 difficulties. 71 difficulties.
72 72
73 \section{Perturbation and Transformation of Character Images} 73 \section{Perturbation and Transformation of Character Images}
74 74
75 \subsection{Adding Slant}
76 In order to mimic a slant effect, we simply shift each row of the image proportionnaly to its height.
77 The coefficient is randomly sampled according to the complexity level and can be negatif or positif with equal probability.
78
79 \subsection{Changing Thickness}
80 To change the thickness of the characters we used morpholigical operators: dilation and erosion~\cite{Haralick87,Serra82}.
81 The basic idea of such transform is, for each pixel, to multiply in the element-wise manner its neighbourhood with a matrix called the structuring element.
82 Then for dilation we remplace the pixel value by the maximum of the result, or the minimum for erosion.
83 This will dilate or erode objects in the image, the strength of the transform only depends on the structuring element.
84 We used ten different structural elements with various shapes (the biggest is $5\times5$).
85 for each image, we radomly sample the operator type (dilation or erosion) and one structural element
86 from a subset depending of the complexity (the higher the complexity, the biggest the structural element can be).
87 Erosion allows only the five smallest structural elements because when the character is too thin it may erase it completly.
88
75 \subsection{Affine Transformations} 89 \subsection{Affine Transformations}
76 \subsection{Adding Slant} 90 We generate an affine transform matrix according to the complexity level, then we apply it directly to the image.
91 This allows to produce scaling, translation, rotation and shearing variances. We took care that the maximum rotation applied
92 to the image is low enough not to confuse classes.
93
77 \subsection{Local Elastic Deformations} 94 \subsection{Local Elastic Deformations}
78 \subsection{Changing Thickness} 95 \subsection{GIMP transformation}
79 \subsection{Occlusion} 96 \subsection{Occlusion}
80 \subsection{Background Images} 97 \subsection{Background Images}
81 \subsection{Salt and Pepper Noise} 98 \subsection{Salt and Pepper Noise}
82 \subsection{Spatially Gaussian Noise} 99 \subsection{Spatially Gaussian Noise}
83 \subsection{Color and Contrast Changes} 100 \subsection{Color and Contrast Changes}
138 155
139 \subsection{Training with More Classes than Necessary} 156 \subsection{Training with More Classes than Necessary}
140 157
141 \section{Conclusions} 158 \section{Conclusions}
142 159
143 \bibliography{strings,ml,aigaion} 160 \bibliography{strings,ml,aigaion,specials}
144 \bibliographystyle{mlapa} 161 \bibliographystyle{mlapa}
145 162
146 \end{document} 163 \end{document}