To achieve non-contact daily mental stress detection, this paper proposes a image photoplethysmography to detect mental stress. First, a video of the subject's face is recorded by the cell phone camera. Then, the proposed Dynamic Region of Interest (ROI) extraction method based on Face Mesh is used to obtain the weak skin color changes caused by heart rate fluctuations. Next, the Fast Independent Component Analysis (FastICA) algorithm, wavelet transform and narrowband bandpass filtering are combined to extract the signal and heart rate variability information based on image photoplethysmography. Then, stress-induced experiments are conducted on 30 subjects to screen 14 features for mental stress detection by comparing the differences in heart rate variability parameters between normal and stressful states, and to explore the relationship between short-term mental stress and daily mental stress due to stress induction. Finally, an additional 67 subjects are tested for daily mental stress, and a triple classifier for mental stress detection is built using the machine learning algorithm. The experimental results show that the accuracy of the three classifications of mental stress can reach 95.2%. Given that this method does not require long-term measurements and can accurately detect human mental stress levels using only a smartphone, and that the measurement method is simple, and easy to administer, and does not affect the normal psychological and mental state of the subject, it can be used as a valid tool in psychological research.