Fusion of Saliency and Edge Map for Multi-operator Image Retargeting Algorithm

Image retargeting algorithms are primarily used for transferring and displaying images on devices with varying aspect ratios and resolutions. High-performance image retargeting algorithms can prevent issues such as distortion, deformation, and artifacts during the image retargeting process, resulting in retargeted images that maintain visual quality comparable to the original image. The key to this type of algorithm is to find a suitable energy function to estimate the importance of each pixel in the image. This paper proposes a new energy function, which combines the saliency map based on U2Net and the adaptive Canny edge map to optimize the gradient-based image importance map. Furthermore, threshold values are assigned to the two importance maps based on the proportion of salient objects in the overall image. This approach ensures that the target image retains the salient information and structural details of the original image while removing the minimum energy seams, thus avoiding image artifacts and distortions. To prevent excessive seam carving from distorting the image, we protect the main structure of the image by combining the seam carving and scaling algorithms and adaptively switching between the two. The experimental results show that, compared with the existing techniques, our method can generate more visually effective target images, and has achieved better performance on SIFTflow and SSMI evaluation indicators.