Image Reconstruction with Initialized Boundaries from Primal Sketch

This paper presents an image reconstruction method using the initialized boundaries from primal sketch. The primal sketch model integrates the sketchable part and non-sketchable part, two modeling schemes to explain the structural components and the textural components in natural images. It is denoted by Markov random field models for texture images that the non-sketchable part , it is very time-consuming in the image reconstruction process, the sketchable part include several kinds of aligned primitives, the primitives have the geometric, photometric and topological information, so it is necessary to replace the textural representation with a simple and rapid method in the image reconstruction process to reduce reconstruction time, heat diffusion is such a method. Utilizing the photometric information in the sketch graph, diffusing the "heat" from each token to the whole image along with an annealing mechanism. The reconstructed image using the presented method could be used in situations such as mobile terminal where picture quality is required low are still acceptable.