Text detection in natural images with hybrid stroke feature transform and high performance deep Convnet computing. (5th April 2019)

• Record Type: Journal Article
• Authors: Vidhyalakshmi, M.; Sudha, S.
• Other Names: Piccialli Francesco guestEditor.; Jeon Gwanggil guestEditor.
• Is Part Of: Concurrency and computation. Volume 33:Number 3(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2019-04-05
• Subjects: deep CNN -- deep learning -- hybrid stroke feature transform -- region classification -- scene text -- text detection; Parallel processing (Electronic computers) -- Periodicals; Parallel computers -- Periodicals; 004.35
• DOI: 10.1002/cpe.5271
• Languages: English
• ISSNs: 1532-0626

Abstract

Summary: Detecting Text in Images is an important step in Scene Text Recognition. It still remains a very difficult task because of the variation in size, fonts, orientation, illumination conditions, and complex backgrounds in image. In this paper, a new method to detect text in natural images with a hybrid technique using MSER and stroke feature transform and feature classification with Deep convolution neural network is proposed. The Candidate character region from the image is extracted with MSER and stroke feature transform. Next, a Deep convolution neural network is used to extract deep high level features and they are fused with fully connected layers to classify features. The proposed method achieves F‐measures of 0.73, 0.886, 0.889, and 0.885 on four benchmark Datasets SVT, ICDAR 2011, ICDAR 2013, and ICDAR 2015, respectively.