Sketching-alike Cross-validation Contour Extraction

Contour extraction refers to the process of accurately marking the outline of a target object in an image.It is an important basic fundamental operation in processing step for many computer vision and image processing algorithms.The extracted object outline information is widely used in image segmentation,object detection,occlusion and depth reasoning,3D reconstruction and other fields.Existing outline extraction methods either require a large amount of user interaction or the effect of extraction is difficult to meet.This paper introduces a contour extraction method that interactively performs sketchingalike cross-validation.The artist depicts an object,usually using a lot of short and straight strokes to outline the edges of the object,and then draws a contour line from these sketched strokes.Inspired by this sketching skill,this paper starts with the user's initial sketch of interaction,then generates a large number of overlapping sets of local contours through dynamic programming.Finally,the global contour is obtained by global fusion based on principal component analysis.The global contour is cross-validated from a large number of overlapping local contours,and this process ensures that it is accurate and effective.In this paper,a real-time interactive contour extraction system is implemented.We compare it with a variety of other methods and a large number of experiments show that our method can extract contours conveniently and accurately.Face contour extraction is a special case of general object contour extraction.The existing face contour extraction method is difficult to extract the curved shape details of the face contour.Combining the key points position of the face alignment method and the outline extraction framework of cross-validation in sketching,this paper proposes a fully automatic facial contour extraction method from local to global.First,use the key points of the face alignment to fit the initialization curve.This paper observes the parabolic features of the local face contours and validates this feature through extensive experiments.Then a parabolic-guided local face contour extraction method is designed based on this feature.Finally,the final face contour is found using the global fusion method based on principal component analysis.In this paper,the continuous face contour dataset is annotated,and the face contour extraction algorithm in this paper is analyzed in detail and compared with a variety of existing methods.This method shows superior performance.