Content-Based Photo Quality Assessment
read more
Citations
AVA: A large-scale database for aesthetic visual analysis
Learning a No-Reference Quality Assessment Model of Enhanced Images With Big Data
No-Reference Image Sharpness Assessment in Autoregressive Parameter Space
A Deep Network Solution for Attention and Aesthetics Aware Photo Cropping
Content-based photo quality assessment
References
Histograms of oriented gradients for human detection
The Pascal Visual Object Classes (VOC) Challenge
A model of saliency-based visual attention for rapid scene analysis
A model of saliency-based visual attention for rapid scene analysis
Single Image Haze Removal Using Dark Channel Prior
Related Papers (5)
Frequently Asked Questions (13)
Q2. What have the authors stated for future works in "Content-based photo quality assessment" ?
The authors will leave the integration of automatic photo categorization and quality assessment as the future work.
Q3. What are the characteristics of professional photography?
For landscape photos, well balanced spatial structure, professional hue composition, and proper lighting are considered as traits of professional photography.
Q4. What is the method to detect subject areas in photos?
The authors adopt a layout based method [9] to segment vertical standing objects, which are treated as subject areas by us, in photos from the categories of “landscape” and “architecture”.
Q5. What are the categories of photos that are manually divided?
Photos are manually divided into seven categories based on photo content: “animal”, “plant”, “static”, “architecture”, “landscape”, “human”, and “night”.
Q6. How is the clarity of face regions calculated?
The clarity of face regions is computed through Fourier transform by measuring ratio of the area of high frequency component area to that of all frequency components.
Q7. What were the two methods used to extract the subject areas?
Wong et al. [20] and Nishiyama et al. [14] used saliency map to extract the subject areas, which were assumed to have higher brightness and contrast than other regions.
Q8. What is the metric for the two types of templates?
Based on theextracted hue composition features, two Gaussian mixture models are separately trained for the two types of templates.
Q9. What are the two types of features that are used to assess photo quality?
Existing methods of assessing photo quality from the aesthetic point of view can be generally classified into using global features and using regional features.
Q10. What features are proposed for the composition of photos?
In this paper, the authors propose content based photo quality assessment together with a set of new subject area detection methods, new global and regional features.
Q11. What is the likelihood ratio of a photo being high-quality or low-quality?
The likelihood ratio P (I|high)/P (I|low) of a photo being high-quality or low-quality can be computed from the Gaussian mixture models and is used for classification.
Q12. What is the difference between the dark channel and the light channel?
Since dark channel is essentially a minimum filter on RGB channels, blurring the image would average the channel values locally and thus increase the response of the minimum filter.
Q13. What is the performance of the proposed face features?
Their proposed face features are very effective for “human” photos and enhanced the best performance (0.78) got by previous features to 0.95.