Recent years have seen increased attention being given to Blood Pressure (BP) monitoring. Among all kinds of measurements, the monitors based on Pulse Transit Time (PTT) have gain plenty of attention due to its continuous and cuffless features. Additionally, several studies proposed a fancy way to estimate photoplethysmography (PPG) signal simply via a regular webcam. Nevertheless, literatures on issues of integrating these two advanced techniques have emerged on a slowly and scattered way. Furthermore, accuracy of BP prediction model based on PTT is often limited due to the lack of data. To address the above-mentioned problems, we proposed an image based BP measurement algorithm using k-nearest neighbor and transfer learning results from MIMICII database to real task. The study also introduces newly defined PTT features which are especially suitable for image based PPG and domain adaptation. Compared with the state-of-The-Art algorithm, root mean square error of SBP evaluation has been reduced from 15.08 to 14.02.