Real-time Camera Interface for Photo Composition
Download as PPTX, PDF0 likes335 views
This document describes a study on a real-time guidance camera interface to enhance photo aesthetic quality for novice photographers. The system uses computer vision techniques to detect the region of interest in a photo, assess how well it follows the rule of thirds composition guideline, and provides real-time feedback to users. A user study found that photos taken with the guidance interface were rated significantly higher in aesthetic quality by experts and workers than a static gridline interface and better conformed to the rule of thirds. The interface was effective in improving photo composition and quality for novice users.
![Choosing One Photography Rule in One Photo
Scenario
• Rule-of-thirds
• 1) important compositional elements should be placed along these
lines or their intersection [1]
• 2) the proportion of the object of interest should be roughly one third of
the total image space [2]
• One person portraiture
6
[1] Peterson, B. F. (2003). Learning to see creatively, Amphoto Press.
[2] Smith, J. T. (1797). Remarks on rural scenery. Nathaniel Smith ancient Print. Cited by the Wikipedia
page about rule of thirds: http://en.wikipedia.org/wiki/Rule_of_thirds (retrieved September 10, 2014)](https://image.slidesharecdn.com/chi15talk-160429043519/85/Real-time-Camera-Interface-for-Photo-Composition-5-320.jpg)
Real-time Camera Interface for Photo Composition
- 1. Real-time Guidance Camera Interface to Enhance Photo Aesthetic Quality Yan Xu1, Joshua Ratcliff1, James Scovell2, Gheric Speiginer3, Ronald Azuma1 1Intel Labs 2Intel Corporation 3Georgia Institute of Technology
- 3. Learning Photography Takes Time 3
- 4. Real-time Guidance for Novice Users 4 A Camera Interface that can • Understand the scene • Know the object of interest • Give concrete guidance Research question: Is the real-time guidance interface an effective way to enhance photos’ aesthetic quality?
- 5. Choosing One Photography Rule in One Photo Scenario • Rule-of-thirds • 1) important compositional elements should be placed along these lines or their intersection [1] • 2) the proportion of the object of interest should be roughly one third of the total image space [2] • One person portraiture 6 [1] Peterson, B. F. (2003). Learning to see creatively, Amphoto Press. [2] Smith, J. T. (1797). Remarks on rural scenery. Nathaniel Smith ancient Print. Cited by the Wikipedia page about rule of thirds: http://en.wikipedia.org/wiki/Rule_of_thirds (retrieved September 10, 2014)
- 6. An Example for Rule-of-thirds 7
- 8. System Components 9 1. Find the region of interest by face detection and foreground segmentation +
- 9. System Components 10 2. Calculate how much does the region of interest follow rule-of-thirds The bitmap mask for calculating the alignment between subject-of-interest and rule-of-thirds
- 10. System Components 11 3. User interface that guides users to move the camera in the space
- 11. User Study
- 12. Procedure 13 • 40 users take portraiture photos of their friend, using our interface and a static grid interface • 24 professional photographers rated the photos • 48 Mechanical Turk raters rated the photos
- 14. Quantitative Results (1) 15 • Photos taken with real-time guidance UI has better aesthetic quality than static gridline UI • Using the two-factor repeated measures ANOVA, we found that expert photographers and Mechanical Turk workers (MT) rated the photos taken by real-time guidance interface to be significantly better than those taken by static gridline interface (expert: F = 7.62, p < .05, η2 partial = .249); MT: F = 20.41, p < .01, η2 partial = .303). Raters Real-time Guidance UI Static Gridline UI Expert photographers M = 43.95 SD = 22.99 M = 40.60 SD = 22.08 Mechanical Turk workers M = 65.98 SD = 21.87 M = 60.92 SD = 22.42
- 15. Quantitative Results (2) 16 • Users follow rule-of-thirds better when they use real-time guidance interface • Users align the subject to the rule-of-thirds grid better with the RG interface than the SG interface (average diff = 31.76 (on a 0-250 scale), p < .05, one tailed paired t-test) • The proportion of human subject’s width is significantly closer to 1/3 when using the RG interface compared to the SG interface (average difference = 6%, p < .01, one tailed paired t-test). Users tend to have smaller subject/image ratios when using the SG interface
- 16. Qualitative Findings - Experts 17 1. FACE: Genuine/natural smile, emotion, eye contact, glass reflection, teeth, skin tone, 2. BODY: Aliveness, Natural poses, hands, sense of movement and action, head/body proportion 4. BACKGROUND: leading lines (and other prominent lines) and vanishing points, distraction, complement the subject or not. 3. AROUND THE SUBJECT: subject saliency, distracting background right next to the subject 5. BIG PICTURE: Composition (balance, camera angle/distance , rule of thirds), Lighting (exposure, evenness, shadows), Color (white balance, saturation, Camera angle 50%10%0% 20% 30% 40%
- 17. Qualitative Findings – Mechanical Turk Raters 18 1. FACE: Smile, facial expression (naturalness, confidence, attractiveness), teeth, skin, hair, mood(more fun, happier, more relaxed), eye contact 2. BODY: Natural poses, ore flattering body shape; less distraction or cut off by other elements; pose, clothes 5. BIG PICTURE: Lighting and shadows, Color (tone, accuracy, saturation, glare, patches, vividness, true to reality, harmony), composition, camera angle and distance, sharpness 4. BACKGROUND: less distraction, realistic or not, broader view and context, leading lines 3. AROUND THE SUBJECT: subject saliency, distracting background right next to the subject, fg/bg harmony 50%10%0% 20% 30% 40%
- 18. Conclusion and Future Work
- 19. Take-away Message 20 • Real-time guidance interface is effective in terms of improving photo aesthetic quality and user’s conformation to photography rules • Understanding photos in both RGB and depth can help us better evaluate photo quality and provide feedback
- 20. New Capabilities of Understanding Photos 21 • Depth • Segmentation • Geometry • Lighting …
- 21. Thank you! Real-time Guidance Camera Interface to Enhance Photo Aesthetic Quality Yan Xu, Joshua Ratcliff, James Scovell, Gheric Speiginer , Ronald Azuma contact: yan.xu@intel.com