INVARIANT SEGMENTATION OF TEXTURE IN IMAGES OF NATURAL SCENE

Abstract

Texture segmentation is an important task which has been studied by many researchers. It is a very difficult task for images containing natural scenes. In these images, texture in a perceptually uniform region can change gradually in scale and orientation due to perspective distortion. On the other hand, neighbouring regions with the same texture but sharp difference in scales or orientations are perceived as distinct regions. Most existing works on texture segmentation deal with only images containing uniform texture i.e. texture that does not vary in scale and orientation. Typically, these images contain juxtaposed plane texture taken from the Brodatz album. The methods described in these works cannot correctly segment images segment images containing non-uniform texture. There are also a number of recent methods containing non-uniform texture. There are also a number of recent methods on invariant classification of texture. However, these methods do not consider the rate of change of texture scale and orientation. Therefore, they are unable to sement images into regions consistent with human perception/ this thesis presents a novel method that performs perceptually consistent segmentation of images containing natural texture. The method can segment images containing non-uniform into regions that are consistent with human perception. Such texture segmentation is indispensable in applications such as computer understanding of natural images, texture-based image retrieval, etc.