Analysis of image sequences.

<p>(A) Original ring image with overlaid grid. (B) Extracted image patch in the Lambert-azimuthal equal-area projection with the four different patch sizes of 2° (light thin box), 15°, 30°, and 60° (dark thick box). The color code is consistently used in the following figures. (C) Sequence of image patches with time as third dimension. (D) Tapered image patch. Before any image analyses were performed, tapering by a sinusoidal window had been applied to avoid boundary effects. (E) Two-dimensional fast Fourier transformation of a single frame; the color code gives the magnitude of the Fourier components with brighter colors indicating larger magnitudes. Spatial frequency is increasing with distance from the figure center. (F) Color-coded power spectra obtained from (E) showing the power for the different orientations and frequencies in the image. (G) Total power per orientation obtained by taking the mean for the rows in (F). (H) Total power per frequency obtained by taking the mean across columns in (F). Linear fit visualizes the 1/f^α characteristic common for natural scenes. (I) Histograms of the goodness of the linear fit to the frequency-dependent power spectra as plotted in (H) for all image patches of our dataset showing the quality of the 1/f^α frequency dependence in our dataset of natural images.