Laguerre Krawtchouk moment invariants feature extraction technique for shape analysis

The present study is concerned with the development of Feature Extraction (FE) based on the Moment Invariant techniques. The invariant properties errors were identified, when the shape is examined under the Rotation, Translation and Scaling (RTS) factors. Basically, the feature vector extracted from the original image and its counterpart variation should have similarities in their values. The feature vectors produced by the Moment Invariant techniques, that represent the images, are used as the input of classification. The performance of the percentage correct for image classification depends on the feature vectors from the image itself. Therefore, this study is motivated to develop a new algorithm based on the Moment Invariant by using the polynomials coefficients in order to reduce the invariant properties errors. The proposed technique is called as the Laguerre Moment Invariant (LGMI). The LGMI has been hybridized with the existing Moment Invariant techniques, the Zhi-Krawtchouk Moment Invariant (ZhiKMI) and Krawtchouk Moment Invariant (KMI). The new hybrid techniques are then, called as the Zhi-Laguerre Moment Invariant (ZhiLGMI) and the Laguerre-Krawtchouk Moment Invariant (LGKMI) techniques, respectively. There are five (5) existing Moment Invariant techniques that have been utilized in this work, namely the ZhiKMI, KMI, Racah-Krawtchouk Moment Invariant (RKMI), Legendre Moment Invariant (LMI) and Tchebichef Moment Invariant (TMI) techniques, which will be used to compare with the new proposed techniques. There are two main stages to examine the performance of the Moment Invariant techniques, namely the intraclass and interclass analysis. For the intraclass analysis, a set of equations has been implemented to identify the best technique between the Moment Invariants techniques based on the smallest value of Total Percentage Mean Absolute Error (TPMAE). Meanwhile, for the interclass analysis, three (3) types of Artificial Neural Network (ANN), namely Multilayer Perceptron (MLP), Simplified Fuzzy ARTMAP (SFAM) and Quality Threshold ARTMAP (QTAM), have been utilized to classify the shape images based on classes. From the intraclass results, it was found that the spatial quantization error is the main cause of the reduced Moment Invariants capability. However, the proposed LGKMI technique was found to be capable of producing the best feature vectors with the smallest value of TPMAE. The LGKMI technique is also able to classify different images with the highest percentage of correct classification with over 90% of accuracy for all the three (3) Neural Networks employed in the interclass analysis. Based on the results obtained from the intraclass and interclass analysis, it can be concluded that the proposed techniques, particularly the LGKMI technique, is found to be the best Moment Invariants technique in representing the shape feature.