Contrast sensitivity functions (CSFs) describe visual stimuli based on their spatial frequency. However, CSF calibration is limited by the size of the sample collection and this remains an open issue. In this study, we propose an approach for calibrating CSFs that is based on the hypothesis that a precise CSF model can accurately predict image quality. Thus, CSF calibration is regarded as the inverse problem of image quality prediction according to our hypothesis. A CSF could be calibrated by optimizing the performance of a CSF-based image quality metric using a database containing images with known quality. Compared with the traditional method, this would reduce the work involved in sample collection dramatically. In the present study, we employed three image databases to optimize some existing CSF models. The experimental results showed that the performance of a three-parameter CSF model was better than that of other models. The results of this study may be helpful in CSF and image quality research.