Fast Scene Text Detection with RT-LoG Operator and CNN

Abstract

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Fast RT-LoG operator for scene text detection

“…The results are compared against the state-of-the-art methods on ICDAR 2017 RRC-MLT dataset. Similar to [32], our system appears in the top results of the literature for F-measure score. It introduces a slight gap of less than 1.5 % error compared to the brute-force RT-LoG operator.…”

“…This information includes the stroke width w, scale-space s (x, y) and the operator response h(x, y). As discussed in [32], these features can drive a grouping method. They can serve in addition for the scale prediction, the background/foreground normalization, and contrast correction of characters before a text verification stage.…”

“…They can serve in addition for the scale prediction, the background/foreground normalization, and contrast correction of characters before a text verification stage. Figure 21 presents the general architecture of the system proposed in [32] where the RT-LoG operator is embedded. As highlighted from that, this system can achieve a strong detection accuracy and is competitive with the top systems of the literature performing with end-to-end CNN models and GPU architectures.…”

“…As highlighted from that, this system can achieve a strong detection accuracy and is competitive with the top systems of the literature performing with end-to-end CNN models and GPU architectures. However, results reported in [32] are obtained while using the brute-force RT-LoG operator. As discussed in Sects.…”

“…4.2 and 4.3, the proposed RT-LoG operator is two to three time faster for an approximately equal performance for detection. Table 7 reports the results from system in [32] while embedding our proposed RT-LoG operator. The results are compared against the state-of-the-art methods on ICDAR 2017 RRC-MLT dataset.…”