Improved classification images with sparse priors in a smooth basis.

摘要: Classification images provide compelling insight into the strategies used by observers in psychophysical tasks. However, because of high-dimensional nature classification and limited quantity trials that can practically be performed, are often too noisy to useful unless denoising adopted. Here we propose a method estimating use sparse priors smooth bases generalized linear models (GLMs). Sparse basis impose assumptions about simplicity observers' internal templates, they naturally generalize commonly methods such as smoothing thresholding. The GLMs this context provides number advantages over classic estimation techniques, including possibility using stimuli with non-Gaussian statistics, natural textures. Using simulations, show our recovers typically less more accurate for smaller than previously published techniques. Finally, have verified efficiency accuracy approach data from human observer.