Super-resolution of dynamic scenes using sampling rate diversity
Super-resolution of dynamic scenes using sampling rate diversity, processes and systems thereof are provided. The method implementing the processes includes: in a first stage, super-resolving secondary low-resolution (LR) images using a set of primary LR images to create LR dictionaries to represent polyphase components (PPCs) of high resolution (HR) patches of images; and, in a second stage, reverting to a single frame super-resolution (SR) applied to each frame which comprises an entire image, using the HR dictionaries extracted from the super-resolved sequence obtain in the first stage.
1. A computer system comprising:
a CPU, a computer readable memory and a computer readable storage media;
program instructions to super-resolve patches of secondary LR frames, in a first stage, into local areas within an entire HR image defined as patches;
program instructions to create a dictionary to represent PPPCs of the local areas within the entire HR image, in the first stage; and
program instructions to super-resolve primary LR frames into their HR counterparts, in a second stage, by super-resolving small patches with the secondary LR frames being discarded,
wherein the program instructions are stored on the computer readable storage media for execution by the CPU via the computer readable memory.
2. The computer system of claim 1 , wherein the program instructions further:
super-resolve the patches of secondary LR frames with use of a dictionary whose atoms are extracted from the primary LR frames; and
extract a patch of predetermined size from one of the patches of the secondary LR frames to super-resolve into a corresponding larger HR patch of images; and
work on smaller patches with more samples than in the patches of the secondary LR frames to estimate representation of only the HR patch of images.
3. The computer system of claim 2 , wherein a primary set of LR frames of a same resolution level as that of PPPCs of the HR image is used to create a dictionary to represent the PPPCs of the HR image.
4. The computer system of claim 2 , wherein the patches of the secondary LR frames provide decimated versions of the PPPCs (DPPCs).
5. The computer system of claim 4 , wherein an estimated HR image is obtained by interlacing estimated PPPCs.
6. The computer system of claim 1 , wherein:
a preliminary version of the dictionary is obtained first by extracting K(2d+1) 2 patches, each of a predetermined size from the primary LR frames;
selecting K=n, and d=x, thereby creating a spatiotemporally local preliminary dictionary with a calculated amount of atoms, wherein
from these atoms, an intermediate dictionary with number of atoms equal to their dimensionality are created by selecting most relevant atoms using a best individual feature (BIF) selection method.
7. The computer system of claim 6 , wherein the program instructions reduce the dimensionality of the dictionary using a minimal-redundancy-maximal-relevance (mRMR) method to obtain a final dictionary.
8. The computer system of claim 7 , wherein the final dictionary is a sub dictionary of a corresponding intermediate dictionary, where its smaller size contributes to reduction of artifacts, and to enhancement of detail.