This paper presents an algorithm for multiscale image segmentation. Towards this, it proposes a new region model, that of a homogenous region surrounded by ramp discontinuities (a scenario usually encountered in real images). This paper presents the analysis of this model, leading to a robust algorithm for detection of ramp discontinuities in the image, and finally segmentation of the image at different photometric scales. The algorithm is further specialized for detecting very thin regions. The final segmentation algorithm can detect regions in the image, with varying photometric scales, sizes and arbitrary geometric properties. The properties of the algorithm are experimentally verified on synthetic and real images.