Javascript must be enabled for the correct page display

Non-linear combination of features for writer identification

Boeijenk, E.L. (2016) Non-linear combination of features for writer identification. Bachelor's Thesis, Artificial Intelligence.

AI_BA_2016_LISETTEBOEIJENK.pdf - Published Version

Download (4MB) | Preview
[img] Text
Toestemming.pdf - Other
Restricted to Backend only

Download (552kB)


Writer identification is very important in pattern recognition. Previous works have shown that combinations of different features result in a higher performance than the individual feature involved in the combinations. Traditionally, features are combined linearly, averaging the distance measurements for writer identification. However, non-linear combination of the feature vectors has not yet been studied thoroughly. In this thesis, a non-linear combination of different features is studied in which the dimensions of features are first reduced using the principal component analysis (PCA) method and then different features are non-linearly combined using a kernel function. The proposed non-linear combined method is evaluated using five different both textural and allographic features on four data sets. Three classical kernel functions are applied to the features, such as the gaussian, sigmoid and polynomial kernels. Experimental results show that the PCA reduced features result in higher performances for higher dimension features and that the non-linear combination of PCA reduced features using kernel functions result in lower performance except for the combination of Local Binary Pattern and Chain Code. Non-linear combination is thus not preferred over linear combination.

Item Type: Thesis (Bachelor's Thesis)
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 15 Feb 2018 08:14
Last Modified: 15 Feb 2018 08:14

Actions (login required)

View Item View Item