Measuring the perceived aesthetic quality of photographic images

Abstract

There have been few studies that empirically assess the perception of aesthetics in photography. This study addressed a few image attributes that were hypothesized to be important to the perception of aesthetics. Thirty consumer photographers evaluated 450 images in a repeated measures factorial experiment and provided artistic quality ratings on a 0 to 100-point scale. The image set balanced main subject size, images with people and without people, and the type of perspective cue. Three first-party scenes supplied by each observer were evaluated by all observers in the study. Study results indicate that aesthetic preferences were mediated by all three of the study variables as a significant threeway interaction. Principal component and cluster analyses provided further insight into other features that impact aesthetic preference. Images in this study formed a continuum of quality ranging from low aesthetic and low technical quality to high aesthetic and high technical quality.

Digital Image Quality Assessment Based on Standard Normal Deviation

“…For this analysis, scenes were grouped according to patterns of artistic ratings of images by observers. For both people and non-people images, principal component explained the largest amount of the total variability non-people = 42.0%, people = 55.8%) [23]. In this research, in order to analyze affective factors by image contents, we made options to select audiences' impression of contents(negative, neutral, positive) and contents' color impression(cool, neutral, warm).…”

Automatic consumer video summarization by audio and visual analysis

“…There has been some recent research on characterizing consumer photographs based on image quality as well as developing predictive algorithms [16,17]. In particular, the work in [17] provided an empirical study where a set of visual features describing various characteristics related to image quality and aesthetic values were used to generate multidimensional feature spaces, on top of which machine learning algorithms were developed to estimate images' aesthetic scales.…”

“…In particular, the work in [17] provided an empirical study where a set of visual features describing various characteristics related to image quality and aesthetic values were used to generate multidimensional feature spaces, on top of which machine learning algorithms were developed to estimate images' aesthetic scales. Their study was based on a consumer photographic image collection [16], containing 450 real consumer photographic images selected from a number of different sources: Flickrr, Kodak Picture of the Day, study observers, and an archive of recently captured consumer image sets. The ground-truth aesthetic values (ranging from 0 to 100) over the 450 images were obtained through a user study from 30 observers.…”

Supporting personal photo storytelling for social albums

“…Hence, we use a regression-based computational image aesthetics algorithm based on [23]. Our system also includes regressionbased computational face aesthetics algorithm, since it has been shown that different image categories would benefit from different aesthetic metrics [22], and the best high level categorization regarding aesthetics is usually obtained by partitioning the set into people and non-people photos 5 [5].…”

“…There has been some research in trying to understand fa- 5 Note that we will consider photos to be people photos if they have at least 1 face detected by the face detection algorithm.…”