Using Bispectrum Invariants to Classify Translated and Rotated Images

Neha ., TIFR


Friday, 18 August 2017, 17:15 to 18:15



In  this talk, we will study the  PhD-work  of Risi  Kondor in  which he demonstrates  how  bispectrum  invariants  can  be  used  to  classify translated and rotated  images. The basic idea  is to map the image in some 2-dimensional space to a unit sphere. Then  a translation or a rotation of the input image is given  by an element of the group $SO(3)$  acting on the (image wrapped about  the) sphere.  Kondor shows that  the bispectrum  of the image  yields a  translation  and rotation  invariant feature  vector, which can be used to classify digits (this work was done at CMI under the guidance of K. V. Subrahmanyam).