On the Reconstruction Performance of Compressed Orthogonal Moments

In this paper, a wavelet-based technique is applied to three moment feature vectors corresponding to three different families of orthogonal moments. The resulted compressed vectors are studied experimentally, in order to extract useful information about their behaviour to a reconstruction procedure. The reconstruction performance of these moments is identical to the amount of image information that they contain to certain moment orders. Since the moment vectors are imposed to compression at the high frequency components, a conclusion about their information redundancy can be also determined. The most efficient moment family, by means of the reconstruction error, will form feature vectors with low dimension, yet with high information content and thus will be very useful for pattern recognition applications, guarantying high recognition rates.