A Lagrangian optimization approach to complexity-constrained TSVQ

We present a new variable rate tree-structured vector quantizer (TSVQ) design algorithm, in which the complexity-distortion tradeoff is explicitly managed using a Lagrangian optimization approach. The algorithm is greedy and uses subvector distortion measures to lower the encoding complexity. We show that we can obtain low complexity encoders for the Gauss-Markov source with similar distortion to that observed on standard variable rate TSVQ.

[1]  David L. Neuhoff,et al.  Quantization , 2022, IEEE Trans. Inf. Theory.

[2]  Lai-Man Po,et al.  Adaptive dimensionality reduction techniques for tree-structured vector quantization , 1994, IEEE Trans. Commun..

[3]  Allen Gersho,et al.  Optimal nonlinear interpolative vector quantization , 1990, IEEE Trans. Commun..

[4]  William A. Pearlman,et al.  Variable-rate tree-structured vector quantizers , 1995, IEEE Trans. Inf. Theory.

[5]  Antonio Ortega,et al.  Rate-complexity-distortion optimization for quadtree-based DCT coding , 2000, Proceedings 2000 International Conference on Image Processing (Cat. No.00CH37101).

[6]  Ulug Bayazit,et al.  Variable-length tree-structured subvector quantization , 1996, Other Conferences.

[7]  Antonio Ortega,et al.  Complexity-scalable transform coding using variable complexity algorithms , 2000, Proceedings DCC 2000. Data Compression Conference.