Performance evaluation of feature-based image alignment techniques for inter-sequence error concealment

Mobile reception of digital TV signals is typically error-prone. In multi-broadcast scenarios, inter-sequence error concealment replaces lost image blocks of distorted TV signals with error-free ones of low-resolution reference signals. As the latter are typically delayed, scaled and possibly cropped, however, spatio-temporal image alignment is crucial. Up to now, spatial alignment parameters have been estimated with moderate computational complexity by matching of the widely-used scale-invariant features. In this paper, more recent features from literature are evaluated to further reduce the complexity of the alignment step. Also, one novel feature transform is proposed. Simulations show that the well-known speeded-up robust features can reduce the complexity by 12%. However, the reconstruction quality is lowered due to the misalignment of single frames. The novel transform achieves a speed gain of 11%, but does not decrease the reconstruction quality significantly.