Abstract: To denoise the severely polluted documental image produced by the process of scanning or duplicating,characteristics of noise and text content are investigated.Gray morphology with constructed four direction structuring elements is applied to filter the severely background noise and retain directional text content information.Local structural tensor is brought in to describe the local gradient difference between noise and text content,making the scale element dynamic,which will be larger when process probable noise area and smaller when it is closed to the text.By this way,denoising is satisfying and easily lost stroke and punctuation are well saved.Finally gray morphologic reconstruction is applied to recover original text content information,picking up image only containing text information.Comparing with normal image processing methods,this proposed approach has got efficient effect,which is suitable to deal with this kind of severely polluted documental image.