This study proposes an improvement of the triangular inequality elimination (TIE) algorithm for vector quantization (VQ). More than 26% additional computation saving is achieved. The proposed approach uses dynamic and intersection (DI) rules to recursively compensate and enhance the TIE algorithm. The dynamic rule changes the reference codeword dynamically and reaches the smallest candidate group. The intersection rule removes redundant codewords from these candidate groups. The DI-TIE approach avoids the over-reliance on continuity of input signal. The VQ-based line spectral pair (LSP) quantization in ITU-T G.729 standard and some standard test images are used to test the contribution of the DI-TIE. Experimental results confirm that the DI rules in the TIE algorithm have an excellent performance. Moreover, in comparison with the quasi-binary search (QBS) approach, both the QBS and the DI-TIE methods are independent on the continuity of input signal. Nevertheless, the DI-TIE approach proposed in the paper is superior to the QBS method in the computation saving issue.
- Image compression
- Speech coding
- Triangular inequality elimination
- Vector quantization