Document image segmentation using Gabor wavelet and kernel-based methods

The document image Segmentation is an important component in the document image understanding. Kernel-based methods have demonstrated excellent performances in a variety of pattern recognition problems. This paper applies kernel-based methods and Gabor wavelet to the document image segmentation. The feature image are derived from Gabor filtered images. Taking the computational complexity into account, we subject the sampled feature image to Spectral Clustering Algorithm (SCA). The clustering results serve as training samples to train a Support Vector Machine (SVM). The initial segmentation is obtained by assigning class labels to pixels of the feature image with the trained SVM. A proper post-processing is used to improve the segmentation result. Several representative document images scanned from popular newspapers and journals are employed to verify the effectiveness of our algorithm.