Texture classification using morphology based features

Abstract

In computer vision or image processing, many algorithms depend on the accuracy of image segmentation stage. In segmentation, different objects or regions in an image are separated. These objects can be classified by their texture into separate classes. Mathematical morphology-based features have been used for classifying textures for many years. Existing methods require size and shape of structuring element which can be different for different set of textures. In this work, we deal with the problem of texture classification using morphology-based features.

In this thesis, we look up different texture classification approaches and propose a novel morphological feature-based method for texture classification. The proposed method weighs the features based on the scale of the texture. This does not require, changing the structuring element for every kind of texture. We have implemented the algorithm and compared it with two other existing approaches using the Keylberg texture database. The preliminary results obtained are encouraging and the proposed method using weighted feature increases the classification accuracy.