|Title||A hybrid algorithm using discrete cosine transform and Gabor filter bank|
|Publication Type||Conference Paper|
|Year of Publication||2006|
|Authors||N Kachouie, N., J. Alirezaie, and P. Fieguth|
|Conference Name||17th Canadian Conference on Electrical and Computer Engineering|
|Keywords||channel bank filters, competitive network training, discrete cosine transform, discrete cosine transforms, eigenvalues and eigenfunctions, eigenvector quantization, feature dimensional reduction, Feature Extraction, filter bank coverage, filter parameters, Gabor filter bank, GDCT, hybrid algorithm, hybrid filter bank, image segmentation, image texture, multiple filter banks, neural nets, neural networks, optimal features, principal component analysis, principal components estimation, texture segmentation, unsupervised learning|
Gabor filters have been widely used for texture segmentation and feature extraction, however there are important considerations regarding filter parameters, filter bank coverage in the frequency domain and feature dimensional reduction. In this paper, a texture segmentation algorithm based on a hybrid filter bank is presented. The proposed method uses a Gabor filter bank and discrete cosine transform (GDCT) to extract the optimal features for texture segmentation. To reduce the feature vector dimension a competitive network is trained to estimate the principal components of the extracted features. The feature vectors composing both Gabor and DCT features are quantized by estimated eigenvectors. The proposed method enables the use of multiple filter banks or larger filter banks consisting of a higher number of channels.