Convolutional Neural Network (CNN) is used in various computer vision fields due to its success. The performance of CNN based algorithm completely depends on the type of input data and how we pre-process it. This paper provides details of various pre-processing techniques, like data augmentation, normalization, resizing, and noise reduction. It also explains their impact on CNN-based image recognition tasks. Through this research, we describe the significance of pre-processing in improving CNN accuracy and robustness.