Curiously Effective Features for Image Quality Prediction
This finding challenges the assumption that sophisticated features are necessary for visual quality prediction, potentially simplifying model design for researchers and practitioners in computer vision.
The paper tackled the problem of predicting image quality by showing that random noise features, when used in highly overparameterized linear regression models, achieve performance on par with models using learned or domain-specific features, with high correlations to human ratings.
The performance of visual quality prediction models is commonly assumed to be closely tied to their ability to capture perceptually relevant image aspects. Models are thus either based on sophisticated feature extractors carefully designed from extensive domain knowledge or optimized through feature learning. In contrast to this, we find feature extractors constructed from random noise to be sufficient to learn a linear regression model whose quality predictions reach high correlations with human visual quality ratings, on par with a model with learned features. We analyze this curious result and show that besides the quality of feature extractors also their quantity plays a crucial role - with top performances only being achieved in highly overparameterized models.