Feature Generation I: Linear Transforms

摘要: This chapter aims the generation of features via linear transforms input samples. A number are presented and reviewed. The basic concept is to transform a given set measurements new features. If suitably chosen, domain can exhibit high “information packing” properties compared with original reasoning behind transform-based that an appropriately chosen exploit remove information redundancies usually exist in samples obtained by measuring devices. For example, image results from devicesuch as X-rays or camera. pixels at various positions have large degree correlation, due internal morphological consistencies real-world images distinguish them noise. one uses features, there will be redundant information. Using Fourier coefficients seems reasonable choice because low-energy, frequency neglected little loss shows just tools palette possible transforms.