This document presents research on using visual features to automatically assess the quality of hotel photos. It introduces visual features like brightness, colorfulness, contrast, sharpness and noisiness that have been shown to influence viewers' emotions. The research evaluates these features on a dataset of over 9,000 hotel photos from TripAdvisor. Results show that photos posted by hotel managers tend to have higher quality than travelers. Photos posted more recently also have higher quality than older photos. Additionally, photos associated with positive reviews have higher quality than negative reviews. The techniques can help with photo quality assessment for research and applications in tourism.
Automatic Hotel Photo Quality Assessment Based on Visual Features
1. ENTER 2018 Research Track Slide Number 1
Automatic Hotel Photo Quality
Assessment Based on Visual Features
Aleksandar Trpkovski1
, Huy Quan Vu1
, Gang Li2
,
Hua Wang1
and Rob Law3
1
Victoria University, Australia
2
Deakin University, Australia
3
Hong Kong Polytechnic University, Hong Kong
Speaker: Huy Quan Vu
Email: huyquan.vu@vu.edu.au
2. ENTER 2018 Research Track Slide Number 2
Outline
• Photo Quality Assessment
• Challenges
• Research Objectives
• Visual Features
• Experiments
3. ENTER 2018 Research Track Slide Number 3
Photo Quality Assessment
• Photos is an important means to transfer destination image to
potential travellers, widely available in travel related websites.
• Photo quality influence customer’s emotion and their decision
making.
– High-quality photos: easier to be remembered by customers.
– Low-quality photos : impact negatively the experience of
potential travelers
• Photo Quality Assessment is an important task for various
applications in tourism:
– Travel Website Evaluation (eg. comparing visual quality)
– Marketing material development (eg. selecting quality photo)
– Traveler’s experience evaluation (eg. emotional experience)
4. ENTER 2018 Research Track Slide Number 4
An Experiment with Hotel Photos
• Based on your own judgment to rank the quality of the
following photos.
Photo A Photo B Photo C
A possible raking: Photo B > Photo A > Photo C
5. ENTER 2018 Research Track Slide Number 5
Challenges
• Photo Quality Assessment is a challenging task for tourism
researchers and practitioners:
– Limited background in photography
– Time consuming for manual assessment
– Impossible for large photo collections
• As a result:
– Existing work in tourism focused on studying photo content, rather
than photo quality.
– Photo evaluation is subjective rather than objective
– Limited understanding about travelers’ emotional experience or photo
taking behavior.
6. ENTER 2018 Research Track Slide Number 6
Research Objectives
• introducing computational techniques based on visual
features for automatic assessment of photo quality.
• evaluating performance of several visual features in case
study on hotel photos:
– Brightness
– Colorfulness
– Contrast
– Sharpness
– Noisiness
• These visual features are selected because they were
found to have influence on viewer’s emotion [Valdez &
Mehrabian, 1994]
7. ENTER 2018 Research Track Slide Number 7
Digital Photo Representation
RBG color space
8. ENTER 2018 Research Track Slide Number 8
Visual features
• Brightness: a measure of the amplitude of color intensity
in digital photo.
– Take a value between 0 and 255. A higher value indicates a
brighter photo. [ITU, 2011]
• Colorfulness: measure the diversity of spectrum
contained in the photo.
– We used colorfulness metric by Hasler and Suesstrunk (2003),
having strong correlation with human score.
– In the range 0 (not colorful) and 109 (extremely colorful)
9. ENTER 2018 Research Track Slide Number 9
Visual features
• Contrast: measure the difference in color and brightness
of an object that makes it distinguishable
– Contrast measure by Pedro and Siersdorfer (2009) takes a value
between 0 and 1 (high value indicates high contrast)
• Sharpness: measure the clarity of detail and edge
definition of a photo (Blanchet & Moisan, 2012).
– Higher indicate better sharpness
• Noisiness: measure random pixel level variation in the
digital images
– Noise level estimation algorithm by Liu et al. (2000). Lower value
for noisiness indicates lower noise level and better photo quality.
10. ENTER 2018 Research Track Slide Number 10
Examples of Visual features
Bri: 152.37 Col: 41.90 Con: 0.64
Shap: 2.06 Noi: 0.04
Bri: 114.95 Col: 15.8 Con:
0.50
Shap: 1.74 Noi: 0.11
11. ENTER 2018 Research Track Slide Number 11
Visual features Evaluation
• Data Collection focus on Melbourne, as a tourism
destination in Australia:
– Hotel Photos and meta data were collected from
Tripadvisor using a web crawler.
– Totally, 9,448 photos for 120 hotels were collected (6,514 photos
were posted by travelers and 2,934 photos were posted by hotel
managers)
• Evaluate the visual features using comparative analysis
and statistical tests:
– Traveler photos vs. Management photos
– Recent vs. Before Photos
– Positive vs. Negative review rating
12. ENTER 2018 Research Track Slide Number 12
Traveler vs. Management
Are photos posted by hotel management of higher quality
than those photos posted by travelers?
13. ENTER 2018 Research Track Slide Number 13
Recent vs. Before
• Photos posted by travelers are divided into 2 groups
– Recent: from 2013 to 2016 (4,774 photos)
– Before: 2012 and before (1,739 photos)
Are photos posted recently of higher quality than those
photos posted by before?
14. ENTER 2018 Research Track Slide Number 14
Positive vs. Negative reviews
• Traveler photos were divided into in groups based on
associated review comments.
– Positive: 3-star rating or more
– Negative: 2-star rating or less
Are photos posted in positive review rating of higher quality
than those photos posted by negative review? rating?
15. ENTER 2018 Research Track Slide Number 15
Summary
• A new approach for automatic photo quality assessment
based on visual features was presented and evaluated.
• Experiment on hotel photos at a tourism destination verified:
– Hotel managers are more likely to post higher quality photos
– Photos posted recently have higher quality than previously in term of
sharpness and noise.
– Travelers who are not satisfied with the hotel tend to post lower
quality photos than those satisfied.
• The presented techniques can be adopted for future research
in photo quality assessment in various tourism context and
applications.
16. ENTER 2018 Research Track Slide Number 16
References
• ITU (2011). Bt.601: Studio encoding parameters of digital television for standard 4:3
and wide screen 16:9 aspect ratios. International Telecommunication Union.
Retrieved on 5 March 2017, from http://www.itu.int/rec/R-REC-BT.601/
• Hasler, D. & Suesstrunk, E. S. (2003). Measuring colourfulness in natural images. In
Proceedings of the SPIE, Vol. 5007, pp. 87-95.
• Pedro, J.S., Siersdorfer, S.: Ranking and classifying attractiveness of photos in
folksonomies. In: Proceedings of the 18th International Conference on World Wide
Web, pp. 771–780. Madrid, Spain (2009)
• Blanchet, G. & Moisan, L. (2012). An explicit sharpness index related to global phase
coherence. In Proceedings of IEEE International Conference on Acoustics, Speech and
Signal Processing, (pp. 1065-1068), Kyoto, Japan.
• Liu, C., Arnett, K.P., Litecky, C.: Design quality of websites for electronic commerce:
fortune 1000 webmasters evaluations. Electron. Markets 10(2), 120–129 (2000)