Research On Seam Carving-based Object Removal And Its Detection

Conventional image sub-sampling,proportional scaling and cropping can not make effective balance between the protection of important content and preservation of overall content.In recent years,content-aware image resizing techniques such as seam carving have been one of the hottest research topics in the field of digital image processing,especially image adaptation.In addition,seam carving can be combined with object labeling technique to remove specific object from digital image or video.However,the seam-carving based object removal is usually time-consuming,and easily leads to visual distortions because too many seams are removed.Moreover,for those images after seam carving,it is difficult to detection those positions where seams are removed.There is still room for the improvement of accuracy and precision for the existing detection algorithms.In this thesis,we researches on the seam-carving based object removal and its corresponding detection techniques.It is based on the original seam carving technique proposed by Isareal scholars such as Avidan.Specifically,the main works are summarized as follows.Firstly,a novel discontinuous seam carving approach is proposed for object removal.It only consideres the energy of those pixels outside the target object region.The energy map is computed in two directions,i.e,from bottom to top and from top to bottom,respectively.Then,we carve the seam respectively from the upper bound of the target region to up,from the lower bound of the target region to down and the middle part within the region.Compared with the object removal by exemplar-based image inpainting,the proposed approach can achieve better visual quality.Secondly,a mini-block analysis approach is adopted for the detection of image seam carving.It converts the candidate test image into its intensity component.Then,the relationships between mini-blocks are used to determine the block types that most likely meet the effects of seam carving.Thus,the detection of image seam carving is achieved.Its performance is compared with the existing approach,which uses the Markov model-based statistical features.The experimental verification is conducted with MATLAB.Experimental results show that the proposed approaches can achieve better results of object removal and its detection.