Text Extraction from Hoardings by Hybrid Model
D. Jayaram1, J. Shiva Sai2, CRK Reddy3, V. Kamakshi Prasad4
1D. Jayaram*, Dept. of MCA CBIT, Hyderabad, India.
2J. Shiva Sai, Dept. of CSE, CBIT, Hyderabad, India Dr. CRK Reddy, Dept. of CSE, CBIT, Hyderabad, India
3Dr. V. Kamakshi , Prasad, Dept. of CSE, JNTUH, Hyderabad, India
Manuscript received on April 05, 2020. | Revised Manuscript received on April 25, 2020. | Manuscript published on April 30, 2020. | PP: 73-80 | Volume-9 Issue-4, April 2020. | Retrieval Number: D6650049420/2020©BEIESP | DOI: 10.35940/ijeat.D6650.049420
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: There are various techniques available to detect and extract the text from hoardings. Still it is a challenging task to detect text from images of various sizes, orientation, illuminations and color. With a view to improve on these, a hybrid method of text extraction and detection is proposed. The proposed method uses a symmetry features like Mutual Magnitude Symmetry (MMS), Mutual Direction Symmetry (MDS) and Gradient Vector Symmetry (GVS) to identify text pixel candidates from natural scenes. The proposed method is tested on different datasets like ICDAR, CUTE 80 and also images from mobile phones. Implementation of MMS, MDS, and GVS methods on above datasets has been carried out. Text extraction from hoardings in ICDAR is giving 74% accuracy, CUTE80 is giving 76% and on mobile images 83% of accuracy is achieved.
Keywords: MDS, MMS, GVS, edge detection, segmentation.