Eigen-Distortions of Hierarchical Representations
This work addresses the challenge of modeling human visual perception for applications in computer vision and neuroscience, though it is incremental as it builds on existing methods like Fisher information and deep neural networks.
The researchers tackled the problem of comparing hierarchical image representations by their ability to explain human perceptual sensitivity, finding that early layers of VGG16 match human perception better than later layers or a 4-stage CNN, but simple models with local gain control trained on distortion ratings provide substantially better predictions.
We develop a method for comparing hierarchical image representations in terms of their ability to explain perceptual sensitivity in humans. Specifically, we utilize Fisher information to establish a model-derived prediction of sensitivity to local perturbations of an image. For a given image, we compute the eigenvectors of the Fisher information matrix with largest and smallest eigenvalues, corresponding to the model-predicted most- and least-noticeable image distortions, respectively. For human subjects, we then measure the amount of each distortion that can be reliably detected when added to the image. We use this method to test the ability of a variety of representations to mimic human perceptual sensitivity. We find that the early layers of VGG16, a deep neural network optimized for object recognition, provide a better match to human perception than later layers, and a better match than a 4-stage convolutional neural network (CNN) trained on a database of human ratings of distorted image quality. On the other hand, we find that simple models of early visual processing, incorporating one or more stages of local gain control, trained on the same database of distortion ratings, provide substantially better predictions of human sensitivity than either the CNN, or any combination of layers of VGG16.