Image recognition via Bayesian likelihood analysis of wavelet coefficients

This study experiments with a Bayesian approach in recognition of images utilizing a joint-form likelihood of wavelet coefficients built from decomposition. Images of handwritten numerals are attacked via the Mallet decomposition algorithm with Daubechies wavelets to extract the feature vectors of coefficients. The model assumes the coefficient vectors by multivariate normal distributions and employs a Bayesian approach for classification based on the joint form of distributions. The results demonstrate marked improvement in recognition performance at the second level of decomposition.