NIMA: Neural Image Assessment

Automatically learned quality assessment for images has recently become a hot topic due to its usefulness in a wide variety of applications, such as evaluating image capture pipelines, storage techniques, and sharing media. Despite the subjective nature of this problem, most existing methods either (1) only predict the mean opinion score provided by data sets, such as AVA, or (2) learn from pairwise comparisons that do not take into account the global ranking of images. In this talk we describe an approach that predicts the distribution of human opinion scores using a convolutional neural network. We show that this approach has the advantage of being significantly simpler than other methods with comparable performance. Then, we address the shortcomings of learning from pairwise comparisons by regularizing the pairwise empirical probabilities with aggregated rankwise probabilities. Our resulting models can be used to score images reliably with high correlation to human perception. Additionally, it can also assist with adaptation and optimization of photo editing/enhancement algorithms in a photographic pipeline. All this is done without a need for a “golden” reference image, consequently allowing for a single-image, semantic- and perceptually-aware, no-reference quality assessment.