Information density and visual activity in dynamic pictures

Đokić, Kristian. (2018). Information density and visual activity in dynamic pictures. PhD Thesis. Filozofski fakultet u Zagrebu, Department of Information Science.
(Poslijediplomski doktorski studij informacijskih i komunikacijskih znanosti) [mentor Lauc, Tomislava].

Preview

PDF (Croatian) Download (3MB) | Preview

Abstract

Due to their large influence on the society visual media have been a subject of research in the last two centuries. New opportunities for analysis of visual media have been created through digitisation and development of information and communication technologies. A phenomenon observed by many authors is constant growth of video content activity. This thesis puts forward the method of visual media activity measurement based on the background subtraction algorithms that are known for computer vision systems. Gibson (1954) defines the term visual activity as the totality of movement of objects and people along a constant background and visual information gained through movement of the observer. The problem of visual activity measurment stems from the fact that it is desirable to include the recipient of the information because the amount of emitted visual information and perceived visual activity are not necessarily in a correlation. The background subtraction algorithms process video content in such manner that output of the processing is a two-colour video content where one colour is a detected movable object, while the other is stationary background. The key concept in this thesis is that it is possible to quantify movement by using a ratio of the number of pixels of the detected movable objects and the total number of pixels in the given video content. Since several dozen different background subtraction algorithms are available, the Limited Capacity Model of Motivated Mediated Message Processing was used in order to select the suitable one. It is a model capable of quantification of video information density. The model itself also includes recipients of visual information. Following the selection of the background subtraction algorithms, the proposed model for visual activity measuring was compared with the existing visual activity measurement model – Cutting’s visual activity index – and a strong correlation of rs = 0.823 was observed. It is obvious that both models take into consideration similar properties of video content. Furthermore, due to properties of the underlying algorithms themselves, the proposed visual activity measurement model based on background subtraction algorithms is significantly less susceptible to noise than Cutting’s visual activity index. That has also been proven by introduction of noise in video content which led to significant increases of the visual activity index – even in cases of comparably low noise levels – while the same was not observed with the background subtraction visual activity index. It is to be expected that the visual activity mesuring model based on background subtraction will be less sensitive to application of varying levels of video content compression than Cutting’s visual activity index. The final testing of the proposed model was performed using two groups of video content. The first group consisted of 50 recipients of the Academy Award for Best Picture presented by the Academy of Motion Picture Arts and Sciences between 1965 and 2014. The second group consisted of recipients of the MTV Best Music Video award presented between 1984 and 2013. The Mann-Kendall Trend Test was used to test background subtraction visual activity index changes concerning the award-winning feature films in the above 5-decade period. It was revealed that the value grew in the period. The same test was applied to test the changes in the award-winning music videos in the above 3-decade period – likewise revealing a growth of the background subtraction visual activity index. As an alternative to the Cutting’s visual activity index, the background subtraction visual activity index shall allow scientists in the field of humanities and social sciences to quantify visual activity which is hitherto described in qualitative terms in their works. A potential opportunity for further development and automation of the Limited Capacity Model of Motivated Mediated Message Processing is also introduced because measurement of certain aspects of information density could be automated.

Actions (login required)

View Item