Browse Prior Art Database

Method for contour-based text localization in images and videos

IP.com Disclosure Number: IPCOM000008509D
Publication Date: 2002-Jun-18

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for contour-based text localization in images and videos. Benefits include improved functionality and improved performance.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 17% of the total text.

Method for contour-based text localization in images and videos

Disclosed is a method for contour-based text localization in images and videos. Benefits include improved functionality and improved performance.

Background

              Text is one of the most important types of information and is the fastest growing medium on the Internet. Textual representation of information enables easy search, content analysis, and processing of large amount of data. Some data representations such as images and videos may contain text in graphical format. Automatic translation of graphical text into textual format provides very useful and easy-to-handle annotations to media data, supporting the efficient and effective search and retrieval of media objects.

              Conventional OCR software packages only work with text on simple background (usually black letters on white background). In contrast, text in media objects can be any color and occurs on top of complex scene objects and backgrounds. To open up the field of text recognition, algorithms for localizing, extracting and segmenting text to black text on white background must be developed.

General description

              The disclosed method localizes text in images and videos. The search for text lines operates on sets of contour information extracted from an image/video and is very fast (see Figure 1). Characteristic features of character contours and text line structures are exploited to find text lines.

              The key elements of the method include:

§         Text filtering algorithm of the extracted contour information

§         Use of the remaining contours to form text line candidates

§         Text filtering algorithm of the contours, which are part of a text line candidate

§         Multi-level integration of the text line candidates into the final text lines

Advantages

              The disclosed method provides advantages, including:

§         Performance comparable to a state-of-the-art reference system at a much lower computational complexity (on average, 4 times faster), enabling real-time processing on desktop CPUs with 64-bit architecture

§         Processing time not completely dependent on image size

§         Efficiency with high-resolution videos where per-pixel analysis schemes would be very slow

§         Vectorized text representation (on detection stage, not during tracking) returned by the algorithm, which may be more convenient for an OCR system

Detailed description

              The disclosed method is a contour-based text localization scheme. The main distinction of this method from other methods, such asneural network-based methods, is that it operates mostly on a contour representation of the image. The topological and structural characteristics of an image are exploited rather than metrical characteristics. Contours are one-dimensional (1D) objects in two dimensions (2D). Their relative positions are analyzed to identify text lines.

              The search for text lines operates on sets of extracted contour information from the image/video, and is very fast. Characteristic features of character contours and text l...