When diffusion coefficients of different components in a mixture are similar, NMR spectra of pure individual components are difficult to be obtained via a diffusion-ordered spectroscopy (DOSY) experiment. Two-dimensional correlation spectroscopy (2D-COS) is used to analyze the data from the DOSY experiment. Through the properties of the systematic absence of cross-peak (SACP) in the 2D asynchronous spectra, spectra of pure components can be obtained even if their diffusion coefficients are similar. However, fluctuations in peak-position and peak-width are often unavoidable in NMR spectra, which makes SACPs unrecognizable. To address the problem, a 2D quotient spectrum is used to identify the masked SACPs. However, undesirable interference peaks due to the fluctuations in peak-position and peak-width are still present when we extract a spectrum of a component by slicing the 2D asynchronous spectrum across the SACP. A genetic algorithm (GA) is used to select a suitable subset of spectra where the diversities of peak-position and peak-width are significantly reduced. Then, we used the selected spectra to construct a refined 2D asynchronous spectrum so that the spectra of pure components with significant attenuated interference can be obtained. The above approach has been proven to be effective on a model system and a real-world example, demonstrating that 2D-COS possesses a bright perspective in the analysis of the bilinear data from DOSY experiments.