In this paper, a tree detection algorithm designed specifically for high-resolution digital imagery of cultivated trees is presented and evaluated. The algorithm falls into the class of post classification methods. However, it refines results of an initial classified image by providing statistical neighbourhood measures for each detected tree. Consequently, it retains only location of target tree spice. We base our algorithm on mathematical morphology skeleton approach. The resulting algorithm increases our ability to accurately locate individual trees; the goal is to incorporate spatial location in the GIS platform. Skeleton approach reduces a set of objects to a set of points that describes their geometric centres. These points provide the means to define and measure a number of statistics neighbourhood-based distances (Su et al., 1997). Hereby, new spatial measures exploiting the k-order neighbouring techniques and mathematical morphology were developed and found to increase the discrimination between certain spectrally similar classes (Zhang and Murayma, 2000). Applied to olive tree fields from a remotely sensed image, we increased the classification accuracy by not only reducing errors of omission and commission in tree detection, but also overshoots and undershoots in crown boundary delineation.