High Speed Gabor Filter using Haar Wavelets

다국어 초록 (Multilingual Abstract)

In this paper, we propose a new way of improving the operation speed of Gabor filters. If the mask is applied to some images, a number of mathmatical operations increase due to complexity. If the mask size is reduced as small as possible in an allowable range, the operation time may decrease. But it would make a problem because the small mask is difficult to represent both frequency and orientation characteristics. Therefore, the operation speed is overcome by substituting multiplications for additions, instead of changing the mask size. First, Haar wavelets are generated in terms of binary data. We divide the mask into two groups, 1’s and -1’s. Thus, Haar wavelets show both frequency and orientation characteristics. Then, by applying ‘Matching Pursuit’, we obtain the correlations between the original Gabor filters and our binary Haar wavelets. The Gabor filter can be described by linear combinations of Haar wavelets. As a result, the binary wavelets give a benefit to the operation speed. However, some trade-off is necessary between accuracy and speed since it may cause some errors.

목차 (Table of Contents)

• Abstract
• 1. Introduction
• 2. Gabor Filter
• 3. Matching Pursuit
• 4. Haar Wavelet
• 5. Experiment
• 6. Conclusion
• References