Local Texture Analysis

In this paper we propose a new method to detect the global scale of images with regular, near regular, or homogenous textures. We define texture ''scale'' as the size of the basic elements (texels or textons) that most frequently occur into the image. We study the distribution of the interest points into the image, at different scale, by using our Keypoint Density Maps (KDMs) tool. A ''mode'' vector is built computing the most frequent values (modes) of the KDMs, at different scales. We observed that the mode vector is quasi linear with the scale. The mode vector is properly subsampled, depending on the scale of observation, and compared with a linear model. Texture scale is estimated as the one which minimizes an error function between the related subsampled vector and the linear model. Results, compared with a state of the art method, are very encouraging.

Deriving an effective facial feature from original face images is a vital step for a successful automatic facial expression recognition. In this paper, we proposed a new feature descriptor, Local Principal Texture Pattern (LPTP), for expression recognition. We compute the LPTP feature, at each pixel, by extracting the principal directions of the local neighborhood, and we code the intensity differences on these directions. The mixture of direction and contrast information makes our descriptor robust against rotation and illumination changes.  Consequently, we represent each expression as a distribution of LPTP codes. Our experiments demonstrate the superiority of the proposed feature, on two facial expression databases, over the existing methods.

"The aim of this paper is to present a new approach for local
texture analysis called LFP (Local Fuzzy Pattern). The
proposed method employs fuzzy numbers to deal with the
uncertainties from gray levels of image pixels, and calculates
the membership degree of pixels in relation to a texture
region. The performance of our approach has been compared
to the local texture descriptor called LBP (Local Binary
Pattern) showing an error rate of 2.5% against 15% for the
LBP considering the same sample size from the Brodatz album."