Cite as:
Gokberk B, Akarun L, Alpaydin E, 2003, "Selection of kernel location, frequency, and orientation parameters of 2-D Gabor wavelets for face recognition" Perception 32 ECVP Abstract Supplement

Selection of kernel location, frequency, and orientation parameters of 2-D Gabor wavelets for face recognition

B Gokberk, L Akarun, E Alpaydin

In local feature-based face-recognition systems, the topographical locations of feature extractors directly affect the discriminative power of a recogniser. Better recognition accuracy can be achieved by the determination of the positions of salient image locations. Most of the facial-feature-selection algorithms in the literature work with two assumptions: one, that the importance of each feature is independent of the other features; and, two, that the kernels should be located at fiducial points. Under these assumptions, one can only get a suboptimal solution. Here, we present a methodology that tries to overcome this problem by relaxing the two assumptions by using a formalism of subset selection problem. In addition to the topographical information, resolution and orientation are also the two important parameters of a feature extractor, since face contains features having different orientation and resolution. Therefore, it is natural to design a feature extractor which takes into account these characteristics of faces. Among several alternatives, 2-D Gabor wavelets are very suitable for this task. The selection of resolution and orientation parameters at specific face locations is also carried out by a supervised sequential floating subset search. This step concludes our two-level feature selection methodology, in which the first step extracts topographical importance, and the second step extracts frequency and orientation information. Experiments on the FERET data set are presented, and the validity of the approach is discussed.

These web-based abstracts are provided for ease of seaching and access, but certain aspects (such as as mathematics) may not appear in their optimum form. For the final published version of this abstract, please see
ECVP 2003 Abstract Supplement (complete) size: 1712 Kb