Basic competitive neural networks as adaptive mechanisms for non-stationary colour quantisation

In this paper we consider the application of two basic Competitive Neural N etworks (CNN) to the adaptive computation of colour representatives on imag e sequences that show non-stationary distributions of pixel colours. The te sted algorithms are the Simple Competitive Learning (SCL) algorithm and the Frequency-Sensitive Competitive Learning (FSCL) algorithm. Both, SCL and F CSL are the simplest adaptive methods based, respectively, on minimising th e distortion and on the search for a uniform quantisation. The aim of this paper is to study several computational properties of these methods when ap plied to non-stationary clustering as adaptive vector quantisation algorith ms. Nonstationary colour quantisation is, therefore, representative of the more general class of non-stationary clustering problems. We expect our res ults to be meaningful for other algorithms that involve either the minimisa tion of the distortion or the search for uniform quantisers. We study exper imentally the effect of the size of the image sample employed in the one-pa ss adaptation, their robustness to initial conditions, and the effect of lo cal versus global scheduling of the learning rate.