Recently there has been a resurgence of interest in the properties of natural images. Several studies have been conducted over large databases, and results ranging from the simplest single pixel intensity joint distribution to statistics of linear filtering of the log contrast have been reported. These statistics are important not only in image compression, in which the theoretical limits of an algorithm are determined by the underlying prior model, but also for the study of sensory processing in biology. This review summarizes previous work on image statistics, beginning with first and second order statistics of the log intensity, continuing with linear transformations as principal components, independent components, and sparse decompositions. The first transformation leads to a Fourier model in which the power spectrum falls inversely as a power of spatial frequency. The other two generates basis functions spatially localized and oriented, similar to the most common multi-scale decompositions. We will finish describing a statistical model for natural photographic images, when decomposed in a multi-scale wavelet basis.