A Prominent Object Size Adjustable Method for Content-aware Image Resizing

Traditionally,the mesh-based image resizing has a limitation that prominent objects in an image will always be reduced or magnified along with the image reducing or enlarging.In this paper,a new measurement model about quad distortion energy by considering both the quad's shape and size is proposed to avoid such disadvantage.Moreover,a significance map is redefined as the weighted average of image gradient and saliency,in which both the structural and perceptual information are considered.Finally,single resolution based visual attention model which is calculated from the rarity of features is improved to be a multi-resolution model.It can combine the merits of that the visual attention model is sensitive to the edge of prominent object in the higher resolution and to the interior region of the prominent object in the lower resolution.Testing with many images,we demonstrate that the proposed approach achieves superior accuracy significance map and can change the primary object's size adapting to user preference while preserves the image prominent features.