To visualize Urdu handwritten numerals, we used tdistributed stochastic neighbor embedding (tSNE) (or digits). The data set used is made up of 28 28 images of handwritten Urdu numerals. Authors from various categories of native Urdu speakers were invited to contribute to the data set. Shape similarity between some of the digits is one of the most difficult and critical issues for correct visualization of Urdu numerals. This problem was solved with tSNE by utilizing the local and global structures of the large data set at various scales. The global structure is made up of geometrical features, whereas the local structure is made up of pixelbased information for each class of Urdu digits. We present a novel method for fusing these two independent spaces using Euclidean pair wise distances in a highly organized and principled manner. The fusion matrix embedded with tSNE assists in locating each data point in a two (or three) dimensional map in a novel way. Furthermore, Our proposed method focuses on preserving the highdimensional data’s local structure while mapping to a lowdimensional plane. Our approach is unique in that we embed Euclidean distances in standard t-SNE in order to successfully visualize high-dimensional data represented by multiple independent observations. On our handwritten Urdu numeral dataset, the visualizations produced by tSNE outperformed other classical techniques such as principal component analysis (PCA) and autoencoders (AE).
Author (S) Details
Mujtaba Husnain
Department of Computer Science & IT, The Islamia University of Bahawalpur, Bahawalpur 63100, Pakistan.
Malik Muhammad Saad Missen
Department of Computer Science & IT, The Islamia University of Bahawalpur, Bahawalpur 63100, Pakistan.
Shahzad Mumtaz
Department of Computer Science & IT, The Islamia University of Bahawalpur, Bahawalpur 63100, Pakistan.
Muhammad Muzzamil Luqman
L3i, La Rochelle University, Avenue Michel C´repeau, 17000 La Rochelle, France.
Mickaël Coustaty
L3i, La Rochelle University, Avenue Michel C´repeau, 17000 La Rochelle, France.
Jean-Marc Ogier
L3i, La Rochelle University, Avenue Michel C´repeau, 17000 La Rochelle, France.
View Book :- https://stm.bookpi.org/CTMCS-V2/article/view/1781
