deletions | additions       diff --git a/chapter_Analysis_label_chap_analysis__.tex b/chapter_Analysis_label_chap_analysis__.tex  index fbad894..f057f12 100644  --- a/chapter_Analysis_label_chap_analysis__.tex  +++ b/chapter_Analysis_label_chap_analysis__.tex 
...
\times10\,m \times 7\,m$, with relatively constant lighting settings  due to artificial lighting.  The choice of the parameters is dependent on the environment and the size of the training dataset. Therefore, there is no general optimal parameter. Instead, the parameters have to be adapted to the particular environment. Since the proposed algorithm is intended for known environments, this is always possible.\section{Analysis -- Determining the Number of Textons}  \label{sec:numtextons}  The developed framework allows to tune the computational complexity by  modifying the number of extracted image patches (samples). To increase  the speed of the algorithm, the goal is to use as few samples as  possible. To determine a suitable number of extracted samples, in this  experiment, the average cosine similarity between $D = 20$ datasets of  histograms is compared. Each dataset consists of $N = 10300$  histograms. The independent variable is the number of extracted image  patches $M$. The histograms were generated using the same images. Due  to the random sampling of the extracted image patches, the histograms  of each datasets will differ. This deviation will be measured using  the cosine similarity. Therefore, each of the $D$ datasets was  compared to all the other $D - 10$ datasets and the average cosine  similarity was determined as well as the standard deviation of the  cosine similarity was measured. Comparing the cosine similarity  between the histograms has the advantage that the number of samples  can be determined independent of a specific task.