How to choose training and testing data for image classification ?

Hi, I have seen lot of papers and got confused on how to choose training and testing data. In the research paper attached, they did a five fold cross validation. Many of my colleagues told me that we need to use cross validation for obtaining the parameters of the classifier and with those obtained parameters. Later on we should not touch our train data and we need to just test it on the test data. But, the authors of this research papers took 5 cross validation of the whole data and calculated accuracy from it. I request you to please help me out.