Vector Quantization with a Growing and Splitting Elastic Net

A new vector quantization method is proposed which generates codebooks incrementally. New vectors are inserted in areas of the input vector space where the quantization error is especially high until the desired number of codebook vectors is reached. A one-dimensional topological neighborhood makes it possible to interpolate new vectors from existing ones. Vectors not contributing to error minimization are removed. After the desired number of vectors is reached, a stochastic approximation phase ne tunes the codebook. The nal quality of the codebooks is exceptional. A comparison with two well-known methods for vector quantization was performed by solving an image compression problem. The results indicate that the new method is signiicantly better than both other approaches.