TextureGrow: Object Recognition and Segmentation with Limit Prior Knowledge
Zhijun Yao, Han Q.-L. · 2011
In this paper we present a new method for automatically visual recognition and semantic segmentation of photographs. Our automatically and rapidly approach based on Cellular Automation. Most of the analysis and description of recognition and segmentation are based on statistical or structural properties of this attribute, most of them need plenty of samples and prior Knowledge. In this paper, within a few evident samples (not too many), we can first get the texture feature of each component and the structures, then select the approximately location randomly of the objects or patches of them, then we use the Cellular Automata algorithm to "grow" based on texture of different objects. The grow progress will stop When texture grow to the boundary. By this steps a new method is found which allow us use very few samples and low lever experience and get a rapidly and accuracy way to recognize and segment objects. We found that this new propose gives competitive results with limited experience and samples.