Modified Seam-Carving Image Resizing with Human Detection

Abstract

This paper proposes a modification version of Seam Carving for image resizing. It mainly uses the human detection approach of HOG (Histograms of Oriented Gradients) to improve the Seam Carving. Our study indicates that the Seam Carving method will remove the small-energy seams from human body and thus yield the structure deformation of human body. The proposed method first employs HOG human detection to locate the people in the photo. The pixel energy of the rectangular window which encloses the people is enhanced by multiplying a Gaussion weighting matrix. The protection of human body part with higher energy avoids the deformation of human body structure by the seam carving.