ePrints.FRI - University of Ljubljana, Faculty of Computer and Information Science

Evaluating photo aesthetics using machine learning

Domen Pogačnik (2012) Evaluating photo aesthetics using machine learning. Prešeren awards for students.

Download (9Mb)


    The objective of this thesis is the identification of the characteristics which in uence the aesthetic appeal of photographs. On the basis of these charac- teristics, we defined calculable features for automatic assessment of photog- raphy aesthetics using machine learning methods. For the purpose of this thesis, we defined and implemented 73 features to analyze various aspects of a photograph. The majority of the features depended on the identifi- cation of a photographic subject. The features were tested on two sets of photographs. For the first experiment, we used photographs from Flickr web portal and manually identified the subjects with the assistance of experienced photographers. The results indicated that automatic aesthetic assessment is feasible, since the use of machine learning methods provided 95 per cent classification accuracy in reference to the overall ratings of photographs (9 per cent improvement relative to comparable research). Because of lengthy manual subject identification process, the second experiment was based on photographs from DPChallenge portal and the subjects were identified with a face detection algorithm. The photographs had been submitted for contests and were nearing artistic photography. Automatic assessment proved to be challenging but still feasible. With machine learning methods, the results showed 75 per cent classification accuracy relative to the user ratings (3 per cent improvement). We specified the features which contribute to a success- ful classification of photographs, analyzed their in uence and interpreted the results. In conclusion, we offer some suggestions for further research.

    Item Type: Thesis (Prešeren awards for students)
    Keywords: photo aesthetics, evaluating photo aesthetics, machine learning
    Number of Pages: 93
    Language of Content: Slovenian
    Mentor / Comentors:
    Name and SurnameIDFunction
    prof. dr. Franc Solina71Mentor
    Link to COBISS: http://www.cobiss.si/scripts/cobiss?command=search&base=51012&select=(ID=9562196)
    Institution: University of Ljubljana
    Department: Faculty of Computer and Information Science
    Item ID: 3698
    Date Deposited: 21 Dec 2016 11:26
    Last Modified: 10 Feb 2017 08:30
    URI: http://eprints.fri.uni-lj.si/id/eprint/3698

    Actions (login required)

    View Item