Harmonic convolutional networks based on discrete cosine transform

Convolutional neural networks (CNNs) learn filters in order to capture local correlation patterns feature space. We propose these as combinations of preset spectral defined by the Discrete Cosine Transform (DCT). Our proposed DCT-based harmonic blocks replace conventional convolutional layers produce partially or fully versions new existing CNN architectures. Using DCT energy compaction properties, we demonstrate how can be efficiently compressed truncating high-frequency information thanks redundancies domain. report extensive experimental validation demonstrating benefits introduction into state-of-the-art models image classification, object detection and semantic segmentation applications.