In this paper, we propose a novel technique to implement an algorithm for unsupervised texture segmentation. Local Binary Patterns (LBP) have been demonstrated to provide a robust and efficient framework for texture segmentation. The distribution of local binary pattern and the contrast measure is used to evaluate the similarity between the adjacent image regions during the segmentation operation. The algorithm uses a new data structure. The existing data structures such as trees and graphs are modified to construct a quadtree and mergegraph. Quadtree is used for splitting the image into blocks of roughly uniform textures and mergegraph is used for merging similar adjacent regions. The algorithm has been tested using mosaic images consisting of various textures, in addition to natural scenes. This approach proved to yield computationally effcient segmentation of images.