SVD based Image Actual Resolution Estimation

Image up-sampling is a fundamental operation in image processing, which enlarges the size of original image. Though the upsampled image may look better, it contains the same information as the original and requires more computation and storage. To accurately determine the actual resolution of the upsampled image is challenging with few previous studies to be investigated. Here, we proposed a method for estimating the actual resolution of an image based on Singular Value Decomposition (SVD). The proposed method generates multi-resolution images by SVD, and find the peak difference among each level image’s eigenvalues, where the level image below the actual resolution cannot keep enough feature information. The approach is model-free and does not rely on any user-defined parameters. We demonstrate its feasibility on a wide variety of datasets. Finally, we show how our method can be utilized to compress images effectively.