Edge Suppression by Gradient Field Transformation Using Cross-Projection Tensors

摘要: We propose a new technique for edge-suppressing operations on images. introduce cross projection tensors to achieve affine transformations of gradient fields. use these tensors, example, remove edges in one image based the edge-information second image. Traditionally, edge suppression is achieved by setting gradients zero thresholds. A common application Retinex problem, where illumination map recovered suppressing reflectance edges, assuming it slowly varying. present class problems edge-suppression can be useful tool. These involve analyzing images same scene under variable illumination. Instead resetting gradients, key idea our approach derive local using and transform field another them. Reconstructed from modified shows suppressed or textures at corresponding locations. All are does not require any global analysis. demonstrate algorithm context several applications such as (a) recovering foreground layer varying illumination, (b) estimating intrinsic non-Lambertian scenes, (c) removing shadows color obtaining map, (d) glass reflections.