Super-Resolution with Registration of Multiple Resolution Components Using Coding Parameters

This paper proposes a super-resolution method with registration of multiple resolution components using coding parameters when super-resolving decoded video image, to increase image quality and processing speed. In the proposed method, multiple resolution components of decoded video image are extracted by wavelet decomposition. The decoded image is then registrated to spatial low-frequency components of multiple resolution components for intra and inter frames. The inter prediction vector in the coding parameters is used in this registration process as registration vector to increase processing speed. Super-resolved components of the decoded image are assigned from spatial high-frequency components using the registration vector. To increase super-resolved image quality, the quantization parameter in the coding parameters is used to determine whether allocation is possible in the assignment process. After all registration and assignment processes, super-resolved image is reconstructed using the decoded image and its assigned super-resolved components by using wavelet composition. Experiments show that the proposed method with the quantization parameter and inter prediction vector improves super-resolved image quality compared to the other methods, and also significantly improves processing speed with slight decrease image quality compared to the proposed method without inter prediction vector which uses the high-precision exhaustive search block matching.