Incremental Interactive Intrinsic Image Decomposition

Intrinsic image decomposition can provide a lot of valuable information for many tasks in computer vision and computer graphics.Intrinsic images separation targets at decomposing an image into reflectance and shading.Unfortunately,there are two unknows and only one known in the decomposition process,which results in a severely ill-posed problem.So more information of input images are needed.However,most of existing decomposition methods bringing in auxiliary information cannot obtain satisfying results on complex scenes,which have apparent limitations.So a novel incremental interactive intrinsic image decomposition approach is proposed in this paper to improve the accuracy of existing decomposition methods.The approach is based on the results of other automatic decomposition methods.We use simple interaction as guidance,including reflectance guidance and shading guidance.The method learns similar pairwise relation which can be applied to propagate the user interactive effect by one-class Support Vector Machine(SVM).In the process of learning,superpixel approach is introduced to accelerate the optimization.In addition,we introduce the assigned term to help to decompose the regions with strong shading.Simultaneously,an assisted decomposition model is adopted to obtain better results.To save better results immediately and avoid repetitive operations,we also presented an incremental method.The problem is formulated as the minimization of a quadratic function which can be solved in closed form.We demonstrate the availability of our approach with kinds of automatic decomposition methods on standard datasets and some natural images.Also,compared to other interactive methods,our algorithm can get comparable and even better results and can be applied more widely.