In this paper, we bring forth a novel approach of video text detection usingFourier-Laplacian filtering in the frequency domain that includes averification technique using Hidden Markov Model (HMM). The proposed approachdeals with the text region appearing not only in horizontal or verticaldirections, but also in any other oblique or curved orientation in the image.Until now only a few methods have been proposed that look into curved textdetection in video frames, wherein lies our novelty. In our approach, we firstapply Fourier-Laplacian transform on the image followed by an idealLaplacian-Gaussian filtering. Thereafter K-means clustering is employed toobtain the asserted text areas depending on a maximum difference map. Next, theobtained connected components (CC) are skeletonized to distinguish various textstrings. Complex components are disintegrated into simpler ones according to ajunction removal algorithm followed by a concatenation performed on possiblecombination of the disjoint skeletons to obtain the corresponding text area.Finally these text hypotheses are verified using HMM-based text/non-textclassification system. False positives are thus eliminated giving us a robusttext detection performance. We have tested our framework in multi-oriented textlines in four scripts, namely, English, Chinese, Devanagari and Bengali, invideo frames and scene texts. The results obtained show that proposed approachsurpasses existing methods on text detection.
展开▼
