The quest to learn image data distributions efficiently, particularly in environments constrained by computational and data resources, is a significant challenge in the field of computer vision. This paper introduces a modified version of the traditional Restricted Boltzmann Machine (RBM), termed the complex-valued RBM (cvRBM), which incorporates complex-valued inputs by utilizing a two-channel approach where the second channel is generated using a Sobel filter. The cvRBM aims to enhance the model’s capacity to handle image data by processing additional edge-detected information alongside original images. We evaluate the efficacy of cvRBM in comparison to the standard RBM on the MNIST dataset, focusing on their ability to reconstruct input data as measured by reconstruction error. The findings suggest that while cvRBM provides a richer representation of image content, it requires careful tuning to match the performance of traditional RBM models.