Visual Aesthetics and User Preference

Aesthetic analysis goes beyond traditional image quality analysis in that it also needs to take into account subjective, social, emotional and semantic dimensions. Recent advances in semantic analysis, social networking services and machine learning enable the development of aesthetic models. This provides challenges both in image features (traditional 'part-of-object' image representations are not necessarily ideal for detecting if an image is 'futuristic' for example), and also in the area of algorithmics (judgements are subjective, information is often partial or incomplete, etc.). When providing users with content suggestions or tools to generate content, information from many subjective sources has to be pooled together to give the user the right content, or right modification, at the right time, often for a very specific and often one-off task.