In this paper we explore local image descriptors fortexture classification. We mainly propose two novel contributions:an effective sampling structure based on Pixel To Patch (PTP)to mimic the retinal sampling pattern; and a novel LocalNeighboring Intensity Relationship Pattern (LNIRP) descriptorto extract texture feature by exploring neighboring gray-scaleproperties. The LNIRP descriptor is extended by using the PTPsampling structure which aims to capture not only micro-patternsbut also macro-patterns, while reducing feature dimensionalityand improving computational efficiency. The proposed descriptorhas advantages of robustness to image rotation, computationalsimplicity, no texton dictionary learning step and training-free.Moreover, the LNIRP descriptor is complementary to the LocalBinary Pattern (LBP) descriptor. Extensive experiments wereconducted on Outex database to evaluate the proposed descriptorand sampling structure. The proposed descriptor can achievesuperior classification performance compared to most of thestate-of-the-art methods, including what we believe to be thebest results reported for Outex, while offering a smallest featuredimension.