Natural Object Recognition Using the Combination of Gaussian Model and Region Growing

A natural object recognition algorithm using the combination of Gaussian model and region growing is presented. Firstly, Visual features of the training images are modeled through Gaussian model, and prior information about spatial location is joined as well. Then the probability value of each region belonging to every class is computed when the regions with high probability values are added to the model and model parameters are updated at the same time. Finally, region growing algorithm is used to obtain the recognition results. The algorithm is tested on the images from Pasadena Houses2000 database including 5 categories of natural objects, e.g. sky, road, house, tree and grass. Experimental results demonstrate the superiority of this algorithm.