DCT-DWT-FFT based method for text detection in underwater images
Document Type
Conference Item
Publication Date
1-1-2022
Abstract
Text detection in underwater images is an open challenge because of the distortions caused by refraction, absorption of light, particles, and variations depending on depth, color, and nature of water. Unlike existing methods aimed at text detection in natural scene images, in this paper, we have proposed a novel method for text detection in underwater images through a new enhancement model. Based on observations that fine details of text in image share with high energy, spatial resolution, and brightness, we consider Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), and Fast Fourier Transform (FFT) for image enhancement to highlight the text features. The enhanced image is fed to a modified Character Region Awareness for Text Detection (CRAFT) model to detect text in underwater images. To explore enhancement methods, we evaluate six combinations of image enhancement techniques, namely, DCT-DWT-FFT, DCT-FFT-DWT, DWT-DCT-FFT, DWT-FFT-DCT, FFT-DCT-DWT, FFT-DWTDCT. Experimental results on our dataset of underwater images and benchmark datasets of natural scene text detection, namely, MSRA-TD500, ICDAR 2019 MLT, ICDAR 2019 ArT, Total-Text, CTW1500, and COCO Text show that the proposed method performs well for both underwater and natural scene images and outperforms the existing methods on all the datasets.
Keywords
Under water images, Text detection in underwater images, Image enhancement, Discrete cosine transform, Wavelet transform, Fourier transform, Modified Character Region Awareness for Text Detection (CRAFT)
Divisions
Computer
Funders
National Natural Science Foundation of China [Grant no. 61721004],National Key Research and Development Program of China [Grant no. 2018AAA0100400],Indian Statistical Institute
Volume
13189
Publisher
SPRINGER INTERNATIONAL PUBLISHING AG
Publisher Location
GEWERBESTRASSE 11, CHAM, CH-6330, SWITZERLAND
Event Title
6th Asian Conference on Pattern Recognition, ACPR 2021
Event Location
Virtual, Online
Event Dates
9-12 November 2021
Event Type
conference