Context based multiscale classification of images

This paper presents an algorithm to segment images into four classes: background, photograph, text and graph. There are two important aspects about the algorithm. The first is that the algorithm takes a multiscale approach, which adaptively classifies an image at different resolutions. The multiscale structure enables accurate classification at class boundaries as well as fast classification overall. The second is that the context information, which is accumulated in the process of classification, is used to improve the classification accuracy