Comparison of two gabor texture descriptor for texture classification

Gabor texture descriptor have gained much attention for different aspects of computer vision and pattern recognition. Recently, on the rayleigh nature of Gabor filter outputs Rayleigh model Gabor texture descriptor is proposed. In this paper, we investigate the performance of these two Gabor texture descriptor in texture classification. We built a texture classification system based on BPNN, and use the corresponding feature vector from traditional Gabor texture descriptor or Rayleigh model one as input of BPNN We use three datasets from the Brodatz album database. For all the three datasets, the original texture images are subdivided into non-overlapping samples of size 32 x 32. 50% of the total samples are used for training and the rest are used for testing. We compare the system training time and recognition accuracy between two Gabor texture descriptor. The experimental results show that, it takes more time when using Rayleigh model Gabor texture descriptor than traditional one, and the traditional Gabor texture descriptor is more accuracy. Rayleigh model Gabor texture descriptor modifies texture descriptor with nearly half the dimensionality and less computational expense, but it lose some performance compared with traditional one.