Hu Moment-based Reduced Feature Vector Analysis

For classification, features are extracted and converted into appropriate feature sets. In this paper, concentration is focused on to reduce the size of the feature vector that is constructed based on seven Hu invariants. The notion of reduced size of Hu moment is really interesting. Since its inception, seven higher-order Hu moments have been employed by many researchers without exploring ‘why seven’, and why not less numbers of moments. In this paper, we analyzed with various feature vector (FV) sets, which are composed of different combinations of Hu moments and rationalized that based on the characteristics of central moments, it is not necessary to employ all the seven moments in every applications. Through this manner, we can reduce the computational cost. Based on various FV sets, it is evident that we can use lower dimensional feature vectors for various methods of action recognition. Our various experimental evaluations provide evidence that up to the first two or three invariants are sufficient for history images and if we consider energy images, then only the 1st invariant for both images seem adequate for satisfactory recognition.