Applying electromagnetism-like mechanism for feature selection
Information Sciences: an International Journal
| Hi-index | 0.00 |
Content based labels, associated with image sequencesin contemporary video indexing methods, can be textual,numerical as well as abstract, including colour-histogramsand motion co-occurrence matrices. Abstractfeatures or indices are not explicitly numeric entities butrather are composed of numeric entities. When multipleabstract features are involved, distance metrics betweenimage sequences need to be weighted. Most featureweighting methods in the literature assume that the spaceis numeric (either discrete or continuous) and so notapplicable to abstract feature weighting. This paperelaborates some feature weighting methods applicable toabstract features and both binary (feature selection) andreal-valued weighting methods are discussed. Theperformance of different feature selection and weightingmethods are provided and a comparative study based onmotion classification experiments is presented.