查看文章

Our submitted program is an implementation of a k-nearest neighbor classifier build using MESSIF framework [1]. This version uses several global image descriptors: Local Binary Patterns (LBP)[2], Haralick Features [3], Color Structure (from the MPEG-7 descriptors)[4], Granulometry-based descriptor [5] and surface description. Almost all descriptors (except for Color Structure descriptor) were implemented with help of i3d image processing library1. The classification process of each image can be divided into several stages:

1) preprocessing 2) k-NN search 3) joining information from all nearest neighbors and inferring of classification estimate 4) combination of partial classification estimates and computing final classification