In event-related potential (ERP) signal classification, identifying relevant local peaks within specific time ranges is crucial for feature extraction and subsequent classification tasks, particularly in studies concerning psychiatric disorders such as schizophrenia. However, ERP data in schizophrenia research often contain numerous small peaks that contribute little to the classification process. Therefore, it is essential to discern and retain only the significant peaks that convey specific features for improved classification outcomes. Recently, a vision-based smoothing algorithm based on the Upscale and Downscale Representation (UDR) technique has demonstrated its effectiveness in preserving prominent peaks{\textquoteright} features while filtering out non-salient peaks from the signal waveform. Under UDR{\textquoteright}s operation, the input signal is visualized in the image domain. The input shape is subjected to a thinning algorithm and the resulting skeleton is projected back to the signal domain. This process is similar to neurologists{\textquoteright} visual inspection of the signal, where prominent peaks are tagged, and insignificant peaks are disregarded for feature extraction. This study applied UDR to a dataset of ERPs recorded in both patients with schizophrenia and matched controls, to evaluate its effectiveness in signal classification. In addition, we analyzed UDR{\textquoteright}s impact on classification accuracy when fewer epochs or fewer ERP channels were used. We tested these effects using several classifiers. Experimental results indicated that EEGNet exhibited the most significant enhancement when UDR was applied across all channels, with an accuracy increase of 2.55%. Furthermore, when the number of signal epochs was halved, UDR contributed to enhancements in 4 out of 7 models, with the ShallowConvNet showing the highest improvement of 2.4%. Notably, the accuracy enhancement is observed on more models when employing UDR on subdatasets formed by signals from the Fz, Cz, and Pz electrode locations only. These findings underscore the promising potential of UDR in enhancing schizophrenia classification accuracy, particularly when applied to datasets focusing on key channels.