Localizing Text in Images and Videos based on Morphology

Mohamed Amine Ben Atitallah (1) , Rostom Kachouri (2) , Hassene Mnif (1)

1 LETI - Laboratoire d'électronique et des technologies de l'Information [Sfax]
2 LIGM - Laboratoire d'Informatique Gaspard-Monge

Many multifaceted images comprise observable text. If the occurrences of this text can be identified, segmented, and recognized automatically, they will be a prized source of high-level semantics; for retrieval and indexing. In this paper, we will propose a novel method for localizing and detecting text in complex images and video frames based on morphology. A morphological Gardient is generated by computing the variance between the dilation and the erosion image. Then the candidate of regions are connected via a morphological closing operation and every text areas are determined used the occurrence of text in each candidate. The identified text regions are localized perfectly via the projection of the text pixels in the morphological Gardient map. This method is sturdy to different position, character size, color and contrast. The updating of the text region between images is also used to minimize the processing time. Tests are realized on divers images to confirm the good efficient of our method.

Mots clés: object detection; object segmentation; morphology Gardient operator; text detection; text segmentation; video processing

Domaines: Traitement du texte et du document

Mohamed Amine Ben Atitallah, Rostom Kachouri, Hassene Mnif. Localizing Text in Images and Videos based on Morphology. International Journal of Recent Technology and Engineering, 2020, ⟨10.35940/ijrte.A2598.059120⟩. ⟨hal-02919771⟩