Boundary detection method based on supervising for small sample size problem

In this paper, we address segmentation of the image with gray and texture measurements together. Combining the filter banks and improved K-Means clustering, the texton is extracted effectively in small samples case. And then, a model used for boundary detection is proposed. This model combines multiple cues, such as gray and texture feature. Proposed model trains parameters using human labeled images and therefore the output of trained model is detected boundary. Finally, we optimize the extracted boundary. The results show that our method not only can accurately detect the boundary but also reduce the time complexity in small samples case compared to the existing method.