A person can quickly grasp the movie genre (drama, comedy, cartoons, etc.) from a poster, regardless of short observation time, clutter and variety of details. Bearing this in mind, it can be assumed that simple properties of a movie poster should play a significant role in automated detection of movie genres. Therefore, visual features based on colors and structural cues are extracted from poster images and used for poster classification into genres. A single movie may belong to more than one genre (class), so the poster classification is a multi-label classification task. To solve the multi-label problem, three different types of classification methods were applied and described in this paper. These are: ML-kNN, RAKEL and Naïve Bayes. ML-kNN and RAKEL methods are directly used on multi-label data. For the Naïve Bayes the task is transformed into multiple single-label classifications. Obtained results are evaluated and compared on a poster dataset using different feature subsets. The dataset contains 6000 posters advertising films classified into 18 genres. The paper gives insights into the properties of the discussed multi-label classification methods and their ability to determine movie genres from posters using low-level visual features.
multi-label classification data transformation method movie poster