Image normalization is a preprocessing step in computer vision that adjusts the range and distribution of pixel values to improve the performance of machine learning models. It enhances the model's ability to learn by providing a consistent scale and reducing the influence of varying lighting conditions and contrasts in different images.