Dynamic texture recognition using local binary pattern

Abstract

Recognition of objects in a visual scene is of particular interest in many computer vision applications To characterize and recognize any object two types of information serve as a primary cue i e the appearance information and the motion information Although motion plays a vital role in the recognition task a high volume of research has been devoted to the study of object recognition using the appearance information alone and a little attention is paid to the motion information newline To analyze the moving scene the motion information must also be utilized effectively along with surface appearance Towards this end a new term named Dynamic Texture has been introduced in recent years newlineDynamic texture belongs to a family of visual phenomena that exhibits spatially repetitive and certain stationary properties in time Dynamic texture has attracted the attention of the computer vision community very recently due to its application in various fields such as face and activity recognition, environmental monitoring surveillance background detection and so on Despite their wide range of applicability the analysis of dynamic texture is a challenging task due to its unknown spatiotemporal extent newlineThis thesis aims to provide the frameworks and algorithms for the representation and recognition of dynamic texture using Local Binary Pattern which is an eminent texture descriptor that combines computational simplicity with high descriptive efficiency newlineVarious LBP based dynamic texture recognition approaches have been proposed However several issues are still open such as an effective blend of motion with appearance features invariance to various translations noise resistance and so on newlineThis thesis mainly emphasizes on the texture contrast and edge features for the effective characterization of the dynamic texture Moreover all proposed approaches compute the mentioned features in both spatial plane and temporal plane to capture both the appearance and motion information of a dynamic texture This thesis also proposes a multiresolution