FFT-Based Method With Near-Matrix Compression