Future Work

The current implementation can handle at most 3 scripts at a time. Future work would involve segmentation of documents having more than three scripts. Using a large set of Gabor filters that span over the entire frequency range is computationally expensive. Hence, a subset of the features must be selected selected and used for segmentation. In the current implementation, the feature selection is not completely automatic and future efforts can be directed towards developing a fully automatic, unsupervised feature selection module. A major chunk of the total time in segmentation is spent in filtering the images. The overall speed of the system could be increased by implementing the convolution operation in hardware.