This paper presents a method for analyzing Japanese Diet deliberations video footage to determine if a speaker reads from a script while speaking. The study focuses on classifying script reading faces as a preprocessing step for future research. Four main classification methods are proposed: (1) using the success or failure of face detection, (2) utilizing head pose angle based on the PnP problem, (3) employing head pose angle based on a regression model, and (4) using an image recognition model. Twelve videos from a Video Retrieval System for Diet Deliberations were used, and 30-s speaking scenes were extracted to evaluate the proposed approaches. The results indicate that approach (4) achieved the highest classification accuracy.