How can we use only cross validation to evaluate the performance of the classifier ?

Heyy,

I thought that we generally use cross validation of training data to find the best parameter selection of C and gamma, and with these C and gamma we evaluate the performance on testing data. But in few research papers, they directly evaluated k-fold cross validation on the whole dataset (training + testing) to report their accuracy.

Can we use cross validation to report our accuracy also ? Or should it be used for only finding the parameters C and gamma ?

 For your reference this is the  paper which used cross validation to report their accuracy. [http://ieeexplore.ieee.org/document/6117511/][1] . You can find the statement in section 3.1.

I request you to help me out.

[1]: http://ieeexplore.ieee.org/document/6117511/