Fourier-Based Image Classification Using CNN

Abstract

Recently, Convolutional Neural Networks (CNNs) have achieved remarkable success in computer vision, image processing and image processing tasks. Traditional CNN models work directly with spatial domain images. On the other hand, images obtained with Fast Fourier Transform (FFT) represent the Frequency domain and provide an advantage in computational cost by reducing potential calculation complexity. This study uses FFT converted images as input to the CNN algorithm to increase image classification and recognition accuracy and investigates the effects of this. The study begins with a comprehensive review of the foundations and features of FFT. It assumes that by converting the input images from the Spatial domain to the Frequency domain, the input image can be learned more efficiently and better results can be achieved in terms of performance by studying the most important features in the Frequency domain. To evaluate the effectiveness of this assumption, CIFAR-10, MNIST-Digits and MNIST-Fashion datasets were used. As a result, it has been shown that FFT-based preprocessing can improve classification accuracy, especially in cases where the datasets contain high-frequency noise, and it has shown different results in different datasets. Therefore, it is thought that the effect of FFT preprocessing varies depending on the datasets.