In video coding, image quality degradation is inevitable due to quantization steps in the encoding process. We had proposed a blind type (no-reference type) PSNR estimation method which uses SVM and natural image statistics. In this paper, as an application of the blind PSNR estimation, we propose a method to predict PSNRs of video sequences before applying video compression by using image features of the original video sequences. As a result, we reveal that we can predict sufficient PSNRs for more than 80% frames within 5dB prediction errors.