Multi-script Identification from Printed Words

In today's multi-script scenario, documents contain page, paragraph, line and up to word level intermixing of different scripts. We need a script recognition approach that can perform well even at the lowest semantically-valid level of words so as to serve as a generic solution. The present paper proposes a combination of Histogram of Oriented Gradients (HoG) and Local Binary Patterns (LBP), extracted over words, to capture the unique and discriminative structural formations of different scripts. Tested over MILE printed-word data set, this concatenated feature descriptor yields a state-of-the-art average recognition accuracy of 97.4% over a set of 11 Indian scripts. In an end-to-end document recognition system it is correct to assume a skew correction unit prior to script identification. Depending on the amount of skew, the skew correction unit can either yield a correctly aligned document or an inverted one. For script identification in such scenarios, we introduce novel modifications over existing HoG and LBP features to propose - Inversion Invariant HoG (II-HoG) and Inversion Invariant LBP (II-LBP) in order to achieve text inversion invariance. Once the script is recognized, script-specific HoG and LBP feature combination can be used to find the text alignment i.e. 0. or 180. for correction. For the MILE database, first-level inversion-invariant script-identification accuracy for 11 script-set is 95.8% (1% gain over the existing best) while the second-level script-specific orientation-detection accuracy is averaged at 97.7%.