deletions | additions       diff --git a/begin_figure_h_begin_center__.tex b/begin_figure_h_begin_center__.tex  index 3af1e87..517fae1 100644  --- a/begin_figure_h_begin_center__.tex  +++ b/begin_figure_h_begin_center__.tex 
...
\begin{figure}[h!]  \begin{center}  \includegraphics[width=0.7\columnwidth]{figures/overview}  \caption{{\label{fig:overview}The figure illustrates  the workflow of the proposed approach. After obtaining images from an initial flight,  the recorded images are stitched together to create an orthomap. The same  images are used to detect and describe their keypoints using  \textsc{Sift}, followed by finding a homography between the keypoints of the flight  images and the orthomap. The center of the homography is used as  $x, y$ coordinate for labeling the training set. Additionally, the  number of detected matches is saved for the corresponding $x, y$  estimate. Each image of the initial flight data set can be used  and a fixed amount of small $6\times6$ pixel image patches can be  extracted. These can be labeled with the nearest texton, using a  distance measure, for example, Euclidean distance. Finally, a  classifier can be trained using texton histogram as feature vector  and the corresponding $x, y$ coordinate as target value. This  process allows shifting computational burden of the \textsc{Sift}  algorithm and homography finding to a faster machine learning  approach.%  }}  \end{center}  \end{figure}