Research On Shadow Detection And Removal Of Single Image

Shadow detection and removal in the field of image processing is a very important basic task that can be used to restore scenes,extract features and detect specific targets.It is widely used in agriculture,meteorology,telecommunications,intelligent transportation,remote sensing,medicine and other fields.At present,the research on shadow detection and removal methods has achieved some results,but the randomness,diversity and complexity of shadow generation have brought certain limitations to the existing algorithms and various algorithms.It is difficult to meet application requirements.Therefore,in-depth research and optimization of various shadow detection and removal algorithms to improve the universality of shadow detection and removal algorithms have important practical significance.This thesis focuses on research and description of shadow detection and removal of a still image.Mainly introduces the principles and methods of image shadow detection and removal dependence;analyzes the types of image shadows,discusses the different properties of shadows in various situations;organizes and elaborate some classic and emerging detection and removal They are classified;a natural shadow image shadow detection algorithm,a removal algorithm,and a removal algorithm for document shadows are proposed.The main work content is as follows:First,for the situation where the accuracy and speed cannot be well balanced in the field of natural shadow detection,a shadow detection algorithm based on HSI color space and OTSU threshold division is proposed.This method can avoid the influence of dark and bright areas on shadow detection,and can still accurately perform shadow detection on different textures when the image has uneven color textures,which can achieve higher detection accuracy and also has lower Complexity and short calculation time.Secondly,for the removal of shadows from natural shadow images,a shadow removal method combining Poisson's equations and penumbra expansion is proposed.The properties of Poisson's equations are used to restore the light inside the shadow.Then,the penumbra expansion method is used to eliminate.Experiments show that this method has a good removal effect,and the processed pictures have no obvious shadow residue traces.Finally,for a special kind of shadow image in natural shadow images: document shadow images,a method of shadow detection and removal is proposed.First,the Gaussian mixture model with the maximum expected fit is used to obtain the initial cluster center through SVM pre-classification,and then the Gaussian mixture model is used for clustering to obtain a shadow mask.Aiming at the problem that the document shadow is greatly affected by the content of the document,the water injection algorithm is used here to obtain the real lighting background.Using the principle of the Lambertian surface model,the shadowless lighting map is finally calculated.Experiments show that this method has a good effect on light restoration,and has no damage to image details,and can obtain a good quality shadowless document image.