The Interestingness of Images

Abstract

We investigate human interest in photos. Based on our own and others' psychological experiments, we identify various cues for "interestingness", namely aesthetics, unusualness and general preferences. For the ranking of retrieved images, interestingness is more appropriate than cues proposed earlier. Interestingness is, for example, correlated with what people believe they will remember. This is opposed to actual memorability, which is uncorrelated to both of them. We introduce a set of features computationally capturing the three main aspects of visual interestingness that we propose and build an interestingness predictor from them. Its performance is shown on three datasets with varying context, reflecting diverse levels of prior knowledge of the viewers.

Interestingness Prediction by Robust Learning to Rank

“…The problem of automatically predicting if people would find an image or video interesting has started to receive increasing attention [7,16,21]. Interestingness prediction has a number of real-world applications.…”

“…First, what defines interestingness and what cues contribute to the human perception of interestingness are still under investigation in psychophysics [39], cognitive sciences [4] and recently computer vision [7,16,21]. Therefore current research in computer vision on interestingness is primarily focused on designing relevant feature representations.…”

“…Different from existing approaches which first remove outliers by majority voting, followed by regression [16] or learning to rank [21], we formulate a unified robust learning to rank framework to jointly solve both the outlier detection and interestingness prediction problems. Critically, instead of detecting outliers locally and independently at each pair by majority voting, our outlier detection method operates globally integrating all local pairwise comparisons together to minimise a cost that corresponds to global inconsistency of ranking order.…”

“…(2) Outliers -it is well known that crowdsourced data are noisy [6,43,30]. Existing approaches [16,21] solve the outlier problem by majority voting which requires multiple comparisons for each pair of data points; but its effectiveness is severely limited by the sparsity of the data.…”

“…In order to obtain more reliable interestingness annotation and thus learn better prediction models, recent studies [16,21] propose to model interestingness from human pairwise comparison data collected using crowdsourcing tools such as Amazon Mechanic Turk (AMT). The annotation task is to select between a pair of images or videos which one is more interesting.…”