Currently, artificial neural networks are advancing tremendously. Especially there are various models related to images and most models receive RGB-based data as Input. RGB-based image data has high resolution and Down-Sampling is required for fitting...
Currently, artificial neural networks are advancing tremendously. Especially there are various models related to images and most models receive RGB-based data as Input. RGB-based image data has high resolution and Down-Sampling is required for fitting input size. However, there is a problem with this method that data loss occurs. To resolve this problem, there are models that extract only important data from image data converted to the frequency domain and set them as input to the model without Down-Sampling process. The problem with these models is that they use a lot of main memory during the preprocessing process that converts RGB-based data into frequency-based data. In this paper, the cost is analyzed at each stage in the data preprocessing process of the frequency domain-based model, and it is shown that the problem of excessive main memory usage in the discrete cosine transform and channel selection process can be improved through suggested preprocessing process based on Kronecker product method.