Motion estimation and compensation in the redundant-wavelet domain using triangle meshes

Abstract

In this paper, a technique is presented that incorporates an irregular triangle mesh into wavelet-domain motion estimation and compensation using a shift-invariant redundant-wavelet transform. The main contribution of this work resides in a demonstration that triangle-mesh motion estimation and compensation can be deployed more effectively in the redundant-wavelet domain thanks to a simple correlation operator that is robust to the prediction residual, a noise-like signal that hinders spatial-domain gradient-based efforts to locate image edges which persist from frame to frame. In the proposed technique, motion compensation takes place through an affine transform mapping triangles from one frame to the next, while the motion-compensated residual is downsampled to a non-redundant form which is then coded using any wavelet-based still-image coder. Experimental results indicate that the proposed technique significantly outperforms not only the usual spatial-domain approach to mesh-based motion estimation and compensation, but also the emerging wavelet-based coders that employ traditional block-based structures, particularly for sequences with fast or complex motion. r

Stereo Video Objects Segmentation Based on Redundant Wavelet Transforms

“…The correlation mask consists of multiplying the high-low (HL) bands, the low-high (LH) bands, and the high-high (HH) bands together and combining the products [2][3] :…”

3D video coding using redundant-wavelet multihypothesis and motion-compensated temporal filtering

“…Motion into subsequent frames is estimated by centering a small block at each vertex in the first frame of the GOF and finding the best matching block in the current frame. We search for the motion of the vertices by minimizing a distortion metric that spans across all subbands of the RDWT decomposition, as we did in [12,13]. Assuming that a GOF contains G frames, this ME process results in G − 1 motion fields, each mapping the first frame of the GOF into one of the other G − 1 frames of the GOF, as illustrated in Fig.…”

Acquisition and analysis of snore signals for diagnosis of obstructive sleep apnea

“…The most likely explanation for the existence of correlation is the inherent spectral features of the noise and the non-orthonormality of TIDWT. This unique interaction property between coefficients and TIDWT is perhaps one of the contributions to the success of waveletdomain motion estimation[141,142]. Here, for denoising applications, we incorporated a correlation limit in the LD threshold to handle this association characteristic between coefficients, with the aim to optimize noise cancellation in translation invariance domain.To determine the correlation limit in the proposed LCD threshold, we examined all the possible correlations in the TIDWT domain.…”