Towards a Computational Approach for the Statistical Style Analysis of Motion Pictures

• Autores: Martin Constable, Wang Junyan, Kap Luk Chan
• Localización: Avanca / Cinema 2012 / Cine Clube de Avanca (dir.), 2012, ISBN 978-989-96858-2-6, págs. 1249-1256
• Idioma: varios idiomas
• Texto completo no disponible (Saber más ...)
• Resumen

  • Statistical Style Analysis of Motion Pictures aims to quantify the style of movies in empirical terms according toprinciples first identified by Barry Salt (B. Salt, 1974). Camera movement type is one of the most important attributesof such an analysis. The frequencies of different types of camera movement in a movie, such as still camera, pan,tilt, track and crane, have been used to determine the style of the movie. Conventionally, the camera movementtype is recognized and annotated manually, which is a very labour-intensive activity. In this paper we propose acomputational approach for recognizing camera movement type for the purpose of semi-automating its statisticalstyle analysis, and we present our preliminary yet promising results. Specifically, based on the tools developed inthe fields of computer vision and pattern recognition, we obtain a semi-automatic software system for classifyingshots into two categories: those with camera movement, and those without. Given a few annotated shots for trainingthe classifier, we are able to classify the rest of the shots automatically. We have achieved up to 90% accuracy inclassifying these two types of shots within a subset of the shots of the movie The English Patient. A by-product of thiswork is a way of visualizing the motion styles of shots within a movie. Outside of Style Analysis, possible applicationof our work would be the automatic or semi-automatic arrangement of clips into sequences, according to stylesderived from our data i.e. Style Transfer.