LoG and Structural Based Arbitrary Oriented Multilingual Text Detection in Images/Video

Introduction

The multimedia data is drastically increasing due to the availability of economical digital cameras. Hence, the multimedia data consists of a massive number of images and videos. Text in an image or a video delivers more detailed and explicit meaning. The textual information is a prominent source with a clear explanation of the content than any other high-level and low-level visual features. Therefore, text present in an image or a video helps us to understand an event or content. There are two types of text present in an image or a video; they are Caption Text: artificial text or superimposed text and Scene Text: Natural existence text captured by the camera. The separation of these texts from the background is challenging and interesting research work in the field of computer vision and pattern recognition.

Text detection is one of the prime part to separate the text from its background. Text detection is a process of determining the existence of text in an image or a video frame. This process does not have any pre-knowledge about textual properties present in an image or a video. The text detection stage is an initial stage of the text information extraction process. Hence it should be a fast and minimum false alarm. The ample of works have done on text detection process, and many of those methods considered particular properties and challenges to distinguish the text region from non-text region. The existing approaches to text detection can be categorized into four techniques, such as edge based, connected component based, texture based, eigenvalue based and region-based methods. But these methods are still facing a problem to detect the text accurately due to illumination, complex background, and variation of size, style, color, orientation and alignment of the text. Arbitrarily oriented text detection is much challenging than horizontal text detection process because the text is present in the form of various direction and text line does not maintain any uniformity. Arbitrary oriented multilingual language is much more difficult than any other text detection process, because the multilingual textual information consists multiples of languages, with different geometrical shapes, multi-color, multi-size, and multi-fonts. Hence, there is a demand for a new technique, which can take care of all the above-discussed problems. The text detection process plays a prominent role in applications like, automatic annotation, indexing, tracking, retrieval, event understanding, assisting tourists and assisting a blind person.