Video Super-Resolution by Motion Compensated Iterative Back-Projection Approach

Traditionally, uniform interpolation based approach is adopted to enhance the image resolution from a single image. Due to the one and only one image, the quality of the reconstructed image is thus constrained. Multiple frames as additional information are utilized to do super-resolution for higher-resolution image. If we have enough low-resolution images with observed sub-pixels, the high-resolution image can be reconstructed. To deal with general cases, we adopted non-uniform interpolation by iterative back-projection to estimate the high resolution image. Motion compensation is used to accurately back-project the kernel and make the process converge efficiently. Motion masks are produced for useful images/regions selection and sub-pixel blocks matching are used to do motion estimation. Objects are assumed to move slightly between two consecutive images. Thus, erroneous motion vectors could be corrected by the center of motion vector clusters. From experimental results, the PSNRs of proposed method were higher than the others, ranging from 0.5 to 1.6 dB. The difference values of the high frequency parts were also greater from 0.63% to 4.86%. It demonstrated the feasibility of the proposed method.