Enhancement of seam carving algorithm for optimized content-aware image resizing application

Seam carving is a popular method of resizing an image while maintaining important content by adding or removing seams. The key to this approach is the importance map, distinguishing seams of the least visual significance. Nevertheless, traditional approaches struggle to preserve local and global image characteristics, causing visible artifacts that can significantly diminish visual fidelity. This study presents a new importance map combining Bubble Entropy, a Euclidean distance saliency map, and Haar wavelets for edge detection. Bubble Entropy is excellent at finding regularity and structural continuity across pixel neighborhoods, tracing significant seams. The saliency map's perceptually uniform property enhances color variation assessment, improving the detection of salient regions and textures. Additionally, Haar wavelet transforms enhance edge detection by maintaining structural transitions while resizing. By calculating the combined importance map, the proposed approach settles at the optimal balance between local and global characteristics of an image. The proposed algorithm outperforms previous seam carving algorithms in terms of perceptual quality metrics: LPIPS, MSSIM, HyperIQA and TOPIQ with the RetargetMe dataset. Execution time was also significantly reduced using the Numba package. Experimental results show image content preservation while improving visual quality and presenting a stronger approach for content-aware image resizing.