=========================== Okazaki Synthetic Texture Image (OSTI) database =========================== Copy right (C) G. Okazawa, S. Tajima, H. Komatsu 2016 This database contains 10355 naturalistic textures (128 x 128 pixels) generated using a texture synthesis algorithm (Portilla & Simoncelli, 2000). Among them, 4170 images use textural parameters extracted from photographs of eight material categories (bark, sand, fabric, fur, leather, stone, water, and wood; an average of 521 images for each). The remaining 6185 images were generated using textural parameters interpolated from the 4170 images. See bellows for the details. Further details can be found in Okazawa et al. (2015). --------------------------- Packages --------------------------- Image_set/00001-10355.bmp : Synthesized texture images Luminance_set/00001-10355.mat : Luminance (cd/m2) of images category.mat : Category of images PSstats.mat : Statistical parameters used to synthesize images FLDA.mat : Dimension reduced version of statistical parameters (see below) PSstats.xlsx : Descriptions of statistical parameters in PSstats.mat mat files can be opened using Matlab(Mathworks) v7 or later. --------------------------- About the packages --------------------------- The database was first developed to measure neural selectivity for naturalistic textures in macaque visual cortices (Okazawa et al., 2015). We collected 4170 photographs of texture surfaces from eight material categories (category.mat), extracted texture synthesis parameters, and projected the parameters onto a low-dimensional space using Fisher's Linear Discriminant Analysis (FLDA). We then created 6185 interpolated textures (the interpolation was performed for the original texture synthesis parameters) to mitigate inhomogeneous distribution of textures on the low-dimensional FLDA space. PSstats.mat contains the texture synthesis parameters of all 10355 textures. FLDA.mat is the parameters of all textures projected on the low-dimensional (7 dimensions) FLDA space. Luminance_set includes actual images synthesized from PSstats.mat using the texture synthesis algorithm. Note that they correspond to luminance values (cd/m2). We converted these luminance data to bmp files (Image_set) assuming gamma = 2.2. --------------------------- Texture synthesis algorithm --------------------------- We used a texture synthesis algorithm developed by J. Portilla & E.P. Simoncelli. Please visit http://www.cns.nyu.edu/~lcv/texture/ for software and further details. --------------------------- Reference --------------------------- To cite the texture database, please use: Okazawa G, Tajima S, Komatsu H (2015) Proceedings of National Academy of Sciences, USA 112:E351-E360 "Image statistics underlying natural texture selectivity of neurons in macaque V4." Since our database is based on the texture synthesis algorithm, please also cite: Portilla J, Simoncelli EP (2000) International journal of computer vision 40:49-70. "A parametric texture model based on joint statistics of complex wavelet coefficients." --------------------------- Contact --------------------------- Gouki Okazawa Center for Neural Science, New York University okazawa@nyu.edu Last update: 11/18/2016