Hybrid Image Retargeting Using Optimized Seam Carving and Scaling
We present a novel hybrid scheme for content-aware image retargeting that allows retargeting images into arbitrary dimensions while preserving visually prominent features and minimizing global information loss. One of the novelties in our scheme is an optimized importance map incorporating the impacts of the gradient map, context-aware saliency map, skin map and Canny edge map. Another novelty is a systematic utilization of both seam carving and scaling for a good balance between information loss and image stretching, where the number of seam operations along each dimension is adaptively determined by a non-linear optimization process. Furthermore, a switching factor is added to the optimization for interactive user control of the switching point between information loss and image stretching. In addition, we propose an optional step to accelerate seam carving by restricting the optimal seam search to a down-sampled thumbnail and the local regions of the input image.
Multimedia Tools and Applications
(2017). Hybrid Image Retargeting Using Optimized Seam Carving and Scaling. Multimedia Tools and Applications, 76(6), 8067-8085.
Available at: https://aquila.usm.edu/fac_pubs/17731