• PDF / 4,206,198 Bytes
• 18 Pages / 595.276 x 790.866 pts Page_size
• 98 Downloads / 208 Views

DOWNLOAD

REPORT

---

ORIGINAL RESEARCH PAPER

Fast RT‑LoG operator for scene text detection Cong Nguyen Dinh1,2   · Mathieu Delalandre2 · Donatello Conte2 · The Anh Pham1 Received: 24 June 2019 / Accepted: 7 January 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract This paper proposes a new real-time Laplacian of Gaussian (RT-LoG) operator for scene text detection. This method takes advantage of the Gaussian kernel distribution in the spatial/scale-space domains and kernel decomposition with the box filtering method. Two levels of optimization are given. The first level of optimization within the spatial domain is obtained by box mutualization. The second level of optimization within the spatial/scale-space domains is performed using a mixed method for box selection. The proposed RT-LoG operator is evaluated on the ICDAR2017 RRC-MLT dataset in terms of robustness and time processing. The results are compared with the state-of-the-art real-time operators for scene text detection. The proposed operator appears as the top performance with the best trade-off between robustness and time processing. The proposed operator can support approximately 30 frames per second (FPS) up to the Quad-HD resolution on a regular CPU architecture with a low-level latency. In addition, the proposed operator can support the full pipeline for scene text detection. Our system is competitive with the top accurate systems of the literature while processing with a difference of two orders of magnitude in term of processing resources. Keywords  Scene text detection · RT-LoG · Stroke model · Box filter · Box selection · Real-time · Predictability

1 Introduction Scene text detection in natural images is an active topic in the image processing and pattern recognition fields. Recent contributions are discussed in surveys [1, 2], and the international contest dedicated to this topic is detailed in [3]. The fundamental and earliest problem investigated in the literature is to make text detection methods robust against variabilities and deformations of text entities in images, which covers different aspects, such as texture and illumination changes, the different scales of characters, the background/ foreground transitions, as shown in Fig. 1.

* Cong Nguyen Dinh [email protected] Mathieu Delalandre mathieu.delalandre@univ‑tours.fr Donatello Conte donatello.conte@univ‑tours.fr The Anh Pham [email protected] 1



Hong Duc University, Thanh Hoa City, Vietnam



Tours University, Tours City, France

2

However, another core problem is to adapt the methods to be time-efficient and real-time, which involves an almost complete reformulation of the methods [4]. The design of real-time methods and systems is a well-known topic in the literature [5]. There are two points, that distinguish the realtime systems from another kind of systems, that are timeliness and predictability. Predictability is related to the design of methods with sharp upper and lower bounds on the execution times. The execution times of methods are guaranteed to prevent