Electroencephalography (EEG) is essential for both diagnosis and monitoring of diseases. Indeed, in the particular context of epilepsy, EEG signals can be significantly affected by the presence of various artifacts. The removal of artifacts from EEG data is crucial as a pre-treatment step for further analysis in the diagnosis of epilepsy. From this application context, several denoising techniques have emerged from EEG signal processing algorithms. In this way, few independent component analysis (ICA) algorithms are used nowadays to process biomedical signals. To this end, this article focuses on the denoising of electroencephalographic signals by canonical correlation analysis (CCA) and by second-order blind source separation (SOBI). Our contribution is based on a synthesis and a comparative study of these two denoising algorithms apply on an epileptic signal. The CCA uses covariance matrices as a descriptor of acquired EEG signals and the SOBI which is based on a joint diagonalization of a set of covariance matrices by exploiting the temporal coherence of the sources. Then we carried out a comparative study between the two algorithms and we emerged the best performing algorithm robustness from the evaluation of the statistical parameters for such a context. The complexity of the processes involved in this field and the lack of reference signals make SOBI a powerful tool for extracting sources of interest according to the results obtained which are satisfactory.