Computational Film Analysis

Am Ende geht die Wahrheit in den Druck. Wie bereits im Genre-Klassiker All the President's Men (Alan J. Pakula, USA 1976) veröffentlicht der investigative Journalist in State of Play (Kevin Macdonald, USA 2019) zuletzt doch noch den Artikel, der die Existenz der Verschwörung zweifelsfrei beweist. Während er nach getaner Arbeit nach Hause geht, kündet der Song «Long As I Can See The Light» der Band Creedance Cleerwater Revival aus dem Off nicht nur vom Aufbruchsgeist einer anderen Epoche, nicht nur vom Nachhausekommen, von einer narrativen Schließung, sondern auch von Licht und Dunkelheit, einem Thema, das leitmotivisch den ganzen Film durchzieht. Das Licht der digitalen Druckerpressen, die nun groß im Bild zu sehen sind, bricht regelrecht in die zuvor konsequent durchgehaltene Dunkelheit des Films ein-nicht zufällig in Form von durchleuchteten Negativen. Denn die Verschwörung wird in Wirklichkeit nicht von der Zeitung, sondern von einem anderen Medium aufgedeckt, einem Medium des Lichts, das nur erscheinen kann, indem es in der Dunkelheit ein Negativ durchleuchtet, eine versteckte Wahrheit ans Licht bringt. Der Film, die Apparatur des Lichts, transzendiert den korrupten politischen Apparat: Das ist die epistemologische Vision des Politthrillers. Politiken des Lichts Die Welten, die dieses Genre entwirft, sind oftmals manichäischer Natur (Taylor 2017, 28). Auf der einen Seite steht die hell erleuchtete Sphäre der Öffentlichkeit, auf der anderen der Raum der Verschwörung, aus dem heraus

We present an ongoing project from the field of quantitative film studies, sometimes also referred to as Cinemetrics (Tsivian, 2009). Most of the related work in this area is focused on quantitative analyses of shot lengths and distributions. In this paper, we suggest color as an additional quantitative parameter for movie analysis and describe an information system that allows scholars to search for movies via their specific color distribution. As a source of condensed movie color information, we make use of the MovieBarcode database. A MovieBarcode is created by skewing each frame of a movie to be only 1 pixel wide. Lining up all these frames in a row creates a barcode-like visualization of the most dominant colors in a movie (fig. 1). Our information system makes use of the color diff 4 library to map more than 1,500

A movie barcode is a graph where the colors used in each frame are extracted and represented in succession. Since it is very intuitive, this tool is mainly used to represent movies and videos artistically and creatively. In this study, we propose the use of movie barcodes with a scientific and colorimetric approach, to summarize in a single image the chromatic variations of a movie. Thanks to this approach, it has been possible to extract movies chromatic mood boards and make comparisons among videos or different versions of the same video. The main potential of this approach is the possibility to study and test different methods of video enhancement and color correction, which could be useful not only in video processing but also in the domain of digital film restoration

The ecological fallacy is an error in the interpretation of statistical data when one ascribes to an individual the qualities of the group to which that individual belongs. In this article I show that using the average shot length (ASL) leads film researchers to commit the ecological fallacy due to the skewed nature of shot length distributions. The ecological fallacy can be avoided when analysing motion picture shot length data using methods that assess the stochastic equality of films’ shot lengths.

I model cutting rates in the late-silent and early-sound films of Alfred Hitchcock as a selfexciting Hawkes process, fitting exponential and power law kernels to the data. The results show that the exponential kernel is a good fit, though it may not be the best fit for every film; and that with the introduction of synchronised sound, the exogeneity of the cutting rate decreased and the endogeneity increased, resulting in slower cutting rates and an increased tendency of shots to cluster.

This paper analyses trailers for films in the Halloween franchise from 1978 to 2018, looking at the structure of the trailers based on segmentation of their soundtracks and the functions of speech, music, and sound effects. The results show that most trailers in the franchise have a similar structure, with approximately two-thirds of the running time devoted to narrative exposition followed by an emotionally intense montage sequence displaying the key attractions of the film's genre. Speech is used to present narrative information to the audience and is gendered, with male characters able to speak with authority; while music and effects serve the dual functions of creating an emotional experience for the audience and sustain a coherent identity for a franchise over time. There is also evidence that relationships exist between trailers within the franchise, either as sequels or remakes, that are relevant to understanding the organisation of a narrative across several films in a franchise.

In this article I analyse motion picture editing as a point process to explore the temporal structure in the timings of cuts in motion pictures, modelling the editing in 134 Hollywood films released between 1935 and 2005 as a Hawkes process with an exponential kernel. The results show that the editing in Hollywood films can be modelled as a Hawkes process and that the conditional intensity function provides a direct description of the instantaneous cutting rate of a film, revealing the structure of a film's editing at a range of scales. The parameters of the exponential kernel show a clear trend over time to a more rapid editing style with an increase in the rate of exogenous events and small increase in the rate of endogenous events. This is consistent with the shift from a classical to an intensified continuity editing style. There are, however, few differences between genres indicating the consistency of editing practices in Hollywood cinema over time and different types of films.

growclusters is an R package that estimates a clustering or partition structure for multivariate data. Estimation is performed under a penalized optimization derived from Bayesian nonparametric formulations. This is done either under a Dirichlet process (DP) mixing measure or a hierarchical DP (HDP) mixing measure in the limit of the global variance (to zero). The latter set-up allows for a collection of dependent, local partitions. This paper revisits the two R Shiny applications that were introduced in our 2019 paper and will focus on additional functionality added to both of them, as well as the integration of the R Shiny applications into the growclusters package. We will also present an example where the R Shiny applications are used to analyze the text from past papers presented at the United Nations Economic Commission for Europe workshops.

This editorial introduces the special issue on “Digital Humanities & Film Studies: Analyzing the Modalities of Moving Images,” which contains a total of eight exciting contributions. Moreover, this editorial aims to highlight the complementary disciplinary perspectives on the computer-aided analysis of moving images, which are important prerequisites to better understand and situate a common DH perspective 1. The Infrastructural Perspective: GLAM Institutions and Film Archives The computer-aided analysis of moving images is becoming an increasingly important topic within the digital humanities and it has been an interdisciplinary field since its beginnings: Computer science and media informatics, film and media studies as well as film archives and documentation centers have developed their own, independent approaches to the digital analysis of film. At the same time, however, several bridges have already been built between the individual disciplines: In computer science, for example...

YouTube is the second most popular website on the internet and a major actor in information propagation, therefore making it efficient as a potential vehicle of misinformation. Current tools available for video platforms tend to hyperfocus on metadata aggregation and neglect the analysis of the actual videos. In an attempt to provide analysts the tools they need to perform various research (behavioral, political analysis, sociology,etc.), we present VTracker (formerly YouTubeTracker), an online analytical tool. Some of the insight analysts can derive from this tool are inorganic behavior detection and algorithmic manipulation. We aim to make the analysis of YouTube content and user behavior accessible not only to information scientists but also communication researchers, journalists, sociologists, and many more. We demonstrate the utility of the tool through some real world data samples.

YouTube is the second most popular website on the internet and a major actor in information propagation, therefore making it efficient as a potential vehicle of misinformation. Current tools available for video platforms tend to hyperfocus on metadata aggregation and neglect the analysis of the actual videos. In an attempt to provide analysts the tools they need to perform various research (behavioral, political analysis, sociology,etc.), we present VTracker (formerly YouTubeTracker), an online analytical tool. Some of the insight analysts can derive from this tool are inorganic behavior detection and algorithmic manipulation. We aim to make the analysis of YouTube content and user behavior accessible not only to information scientists but also communication researchers, journalists, sociologists, and many more. We demonstrate the utility of the tool through some real world data samples.

Computational Film Analysis with R is a first course in the application of computational methods to questions of form and style in the cinema using the R programming language. It aimed at those who have no previous experience of computational film analysis and no prior knowledge of statistics, data science or programming with R is required or assumed. Each chapter discusses the underlying methodological concepts in depth and demonstrates the implementation with R of a range of approaches for analysing sound, colour, editing, cinematography, and scripts of motion pictures, so that after reading this book you will be able to design and execute your own computational film analysis research projects.

This article presents a new data set comprising audio, colour, motion, and shot length data of trailers for the fifty highest grossing horror films at the US box office from 2011 to 2015. This data set is one of the few available for computational film analysis that includes data on multiple elements of film style and is the only existing data set for motion picture trailers suitable for formal analyses. Data is stored in csv files available under a Creative Commons Attribution 4.0 International license on Zenodo: www.doi.org/10.5281/zenodo.4479068.

*** Explore the trailers colour data and create your own visualisations using the interactive Shiny app for this paper at https://nickredfern.shinyapps.io/colour_palletes_US_trailers/ ***


This article analyses the smoothed movie barcodes of 173 trailers nominated for a Golden Trailer award between 2016 and 2019 across nine genres: action, animation/family, comedy, documentary, drama, fantasy/adventure, horror, romance, and thriller. The results show that colours in the nine genres are similar, dominated by dark, unsaturated colours in the orange (hue = 30° ± 30°) and azure (hue = 210° ± 30°) regions of the HSL colour wheel. Colour palettes for each genre have similar colours but show some differences in the diversity and evenness of the distributions of these colours.

I present results showing how the editing style in the films of Alfred Hitchcock changed with the introduction of synchronised sound. I show that the distribution of shot lengths in is well fitted by the three-parameter Dagum distribution and that changes in editing style are captured by the parameters of the fitted distribution.

In this article I introduce MovieBarcodeAnalyzeR, an application for extracting and visualizing colour data in movie barcodes implemented using the shiny framework for the R statistical programming language R. **Now available as an executable file for Windows**.

The viewer's processes of inference making in the cinema involve the framing of hypotheses about the world of the narrative that may be overturned by subsequent information and are, therefore, nonmonotonic. The goal of narrative researchers is to understand the nature of those processes and how texts organise the deployment of those processes in order to present a narrative successfully. To do this we need methods capable of describing processes of hypothesis framing and belief revision. In this paper, I describe the application of the Transferable Belief Model to a hypothetical example of narrative comprehension based on an episode of CSI: Crime Scene Investigation as one such method.

This article refocuses attention onto cutting rates in motion pictures, which have largely been ignored in quantitative studies of film style in favour of shot lengths. I clarify the concept of a rate in statistics and define two equivalent versions of cutting rate in motion pictures. I show that the introduction of the average shot length as a statistical film style was based on a flawed understanding of statistical theory and that it is not a measure of cutting rate, with the result that the tendency of researchers to quote the average shot length as a measure of rate leads to flawed analyses of film style. I demonstrate some simple methods for visualising and analysing cutting rates illustrated by concrete examples for data collected from slasher films.

Computational film analysis combines statistics, data science, information visualization, and computer science with the analytical approaches and domain knowledge of the study of film to frame and answer questions about form and style in the cinema. In this article, I illustrate the visualisation and analysis of the soundtrack of the trailer for the Korean horror film Into the Mirror (2003). Using the spectrogram and normalized aggregated power envelope to show how the sound design of the trailer evolves over time and functions at different scales. I address a range of practical considerations relevant to the methods demonstrated here relating to the use of mono versus stereo audio files, the use of different sampling rates, and audio normalization. A supporting website with the code for the statistical programming language R used in this analysis is available at https://rpubs.com/nr62_rp33/CFA-into-the-mirror.

This article illustrates the use of quantiles as a means of describing and comparing motion picture shot length distributions. This approach is conceptually and computationally simple and leads us to think distributionally about shot lengths rather than focusing on individual values. The result is a better understanding of how this element of film style of two (or more) films differs.

In this article, I analyse the soundtrack of the green band trailer for Sinister (Scott Derrickson, 2012), combining quantitative methods to analyse the soundtrack with formal analysis. I show that, even though Sinister is a narrative about a demon who lives in images, the horror in the soundtrack of this trailer is articulated through the sound design. I describe the structure of the soundtrack and analyse the distribution and organisation of dialogue, the use of different types of sound effects to create a connection between the viewer and the characters onscreen, as well as the use of specific localised sound events to organise attention and to frighten the viewer. I identify two features not previously discussed in relation to quantitative analysis of film soundtracks: an affective event based on reactions to a stimulus and the presence of nonlinear features in the sound envelopes of localised affective events. The sound design of this trailer is consistent with the principles of contemporary sound design in horror cinema, but also demonstrates some variation in its use of sound as a paratext to its parent film.

I analyse the dynamic editing structure of Friday the Thirteenth (1980) using the time series of the ranks of the shot lengths as an exploratory data analytical method for revealing the editing structure of motion pictures without requiring a priori assumptions about the functions of film style. I find that Friday the Thirteenth comprises multiple editing regimes with change points between editing patterns occurring with large changes in mood and localised clusters of shorter and longer takes are associated with specific narrative events. The multiple editing regimes create different types of frightening experiences for the viewer with slower edited passages creating a pervading sense of foreboding and rapid editing linked to the frenzied violence of body horror.

In this paper I analyse the structure of sound in a short horror film Behold the Noose (Jamie Brooks, 2014), applying quantitative methods using open source statistical software in order to determine the structure of the film’s soundtrack and to understand how sounds at different frequencies are used to create an affective experience for the viewer. Examination of the time contour plot the normalised aggregated power envelope shows the soundtrack is organised at different scales, from local transient events, medium scale sections easily distinguishable from those that precede and follow it, and a step change between large scale segments. In this film, sections of increasing emotional attention are associated with on-linear increases in sound energy. Using the short-time Fourier transform I show that sounds at different frequencies are used to create different emotional effects in the viewer; that the range of frequencies used increases as the emotional effect of the soundtrack intensifies; and, that infrasonic frequencies are deliberately and purposefully used as a part of the soundtrack.

The quantitative study of editing patterns in silent film, for the period 1908--1915, has often relied on the average shot length (ASL). It is, for example, clear that the rapidity of D.W Griffith's cutting increased over the period 1908 to 1913 as measured by the ASL. Griffith was a mentor of Mack Sennett whose early directorial work at Biograph emulated Griffith in terms of cutting-rates. When Sennett moved to Keystone there was a marked increase in the cutting-rates employed in the films he directed - this paper suggests that it is possible to trace evolution in his cutting-rates over the period 1912-1914. Sennett was, in turn, a mentor of Charlie Chaplin whose directorial efforts at Keystone have been argued to represent a reaction against the fast cutting expected by that studio. A comparison of the ASLs of the films of Sennett and Chaplin supports this argument, but in certain respects the ASL is a blunt tool for comparative purposes, and more nuanced quantitative analysis is possible. This paper examines some graphical approaches based on the use of cumulative frequency distributions of shot-lengths (SLs) that can be more informative than the use of the ASL alone. Among the graphical techniques illustraed are the averaging of cumulative SLs across bodies of films, using a logarithmic scale to highlight differences, and correspondence analysis of the cumulative SLs to investigatee patterns of difference between individual films. Apart from the suggestion of evolution in Sennett's cutting-rates from late-1912 to 1914 there is greater complexity in Chaplin's practices at Keystone than can be summarized using the ASL. His practice evolved away from the `style' employed by other Keystone directors, in the direction of making greater use of shots of longer duration. However -- and this cannot be inferred from the ASL -- he also made greater use of shots of short duration. That is, to say that he ended his career at Keystone cutting more slowly than Sennett -- based on analysis of ASLs -- oversimplifies things; he was cutting with greater variety as well.