Comparative results of Zernike moments and Pseudo-Zernike moments

摘要

Zernike moments and Pseudo-Zernike moments both are robust to noise and rotational invariant. The Zernike and Pseudo-Zernike moments have been used as a feature extractor for image analysis in various applications. The use of these moments for the recognition is due to their rotation invariance. If we look at the literature, the use of Zernike moments as feature extractor is more as compared to Pseudo-Zernike moments. However, Pseudo-Zernike moments are less sensitive to noise as compared to Pseudo-Zernike moments. This is due to the fact that Pseudo-Zernike moments are more computationally costly as compared to Zernike moments. This paper covers various theatrical and practical issues of both these moments. The comparison results of the recognition of pre-segmented hand-printed Devanagari characters using Zernike and Pseudo-Zernike moments are given. The comparative time taken to compute both Zernike and Pseudo-Zernike from order 0 to maximum order n is also investigated. The performance is investigated on both noisy as well noise-less images. We have conducted experiments with 500 characters/class and the study has been performed on 39 classes. The classifier used is multi-layer perceptron.