In the image processing literature, texture is usually defined in terms of the spatial interactions between pixel values. The aim of texture analysis is to capture the visual characteristics of texture in an analytical form by mathematically modeling these spatial interactions. This allows segmentation of an image into its various textural components, with each component being classified according to how well it fits the mathematical model of a particular texture. This approach requires the number and type of training data sets are used to formalize the criteria by which the texture models become unique from each other, but not necessarily unique from any other textures not included in the training set. If a texture is to be recognized in a scene containing previously unseen textures, then a new approach is required. The texture models need to capture more than just the characteristics required to distinguish one texture from other known textures they need to capture all the unique characteristics of that texture. This paper describes a new method for image segmentation by generating binary random values in the image based on neighborhood spanning tree. This method has produced better result than conventional region based segmentation methods for complex multi resolution images.