Since there are as many ways of preprocessing EEG data as there are ERP studies there is no simple answer to this question. There are a number of EEG preprocessing / ERP calculation software and toolboxes available. But the choices involved in the processing pipeline from raw EEG data to single subject / single trials ERPs is a non-trivial matter and there are many informed choices to be made. I would suggest that you start looking at tutorials of the available software / toolboxes, such as eeglab (https://sccn.ucsd.edu/eeglab/index.php) or Fieldtrip (http://www.fieldtriptoolbox.org/), both Matlab-based, or, if you prefer a python-based option, MNE (https://martinos.org/mne/stable/index.html).

Mauro Ramos Pereira

So, after we obtain the final ERP waveform from filtering, only then can we obtain the ERP components like I mentioned above (N170, P300 etc) ? If so, what is the procedure ? Please help me be clear about it. Thank you

Other often used softwares for EEG preprocessing are EEGlab (https://sccn.ucsd.edu/~scott/ica.html) and Brain Vision Analyzer (BVA) (https://www.brainproducts.com/downloads.php?kid=9). The first one is free (as long as you have Matlab), but a bit less "friendly" in terms of its GUI. A great and easy tool is BVA, but it's unfortunately quite expensive.

What I normally follow as pre-processing steps are:

1) Reference - your online and offline reference may be different, and different options can be used in the process. Some are referring the signal to electrodes on mastoids, some to Cz, some to the average of all the electrodes.

2) Filters - you probably already used some filters during data collection. Often, online filters are less restrict than what you may want to apply in the offline pre-processing. Consider something like 0.1 Hz high-pass, and 40 Hz low-pass filters.

3) Segmentation (or epoching, depending on software) - "cutting" the signal to pieces depending on your time-windows of interest and triggers (or markers, again - a choice of terminology). For example, if your trial consists of a visual stimulus presented for 1000ms and followed by 500ms ITI, you may want to "cut out" segments of -100ms to 1000ms from the onset of the stimulus (marked by a trigger).

4) Baseline correction - the "-100ms" from the previous point will serve now as baseline for the segments. Of course, you can consider a different time window for your baseline. 100 ms is quite standard, though.

5) Blink correction - this can be done in many ways. ICA (Independent Component Analysis) is often used for this purpose. There are plenty of sources online about how it works, and many softwares have implemented algorithms for using ICA on your data.

6) Artifact rejection - blink correction is already a part of artifact rejection, but there's more to this category. You may want to go through your signal and remove all the other artifacts (like movements, floating electrodes, sweat, etc).

7) Averaging - you may want to average the signal for each participant across conditions. This reduces noise. Or you may want to get into single trial EEG (again, many sources online).

8) Choosing ROIs - if you're planning to choose a Region of Interest, you can decide on a few electrodes that would serve as such. Again, there are different ways of how to make these decisions. Often people look at the topographies of their signal and compare that to the previous literature. Deciding on your ROI, you can average the signal from these electrodes and you end up with a datapoint per condition per participant.

It's a long way with many decisions to make, so reward yourself in the end of the process ;)

Good luck!

Thank you everyone for your suggestions.