Generalized text localization in images

Abstract

In some embodiments, the invention includes a method for locating text in digital images. The method includes scaling a digital image into images of multiple resolutions and classifying whether pixels in the multiple resolutions are part of a text region. The method also includes integrating scales to create a scale integration saliency map and using the saliency map to create initial text bounding boxes through expanding the boxes from rectangles of pixels including at least one pixel to include groups of at least one pixel adjacent to the rectangles, wherein the groups have a particular relationship to a first threshold. The initial text bounding boxes are consolidated. In other embodiments, a method includes classifying whether pixels are part of a text region, creating initial text bounding boxes, and consolidating the initial text bounding boxes, wherein the consolidating includes creating horizontal projection profiles having adaptive thresholds and vertical projection profiles having adaptive thresholds.

Description

[0001] 1. Technical Field of the Invention[0002] The present invention relates generally to localization and / or segmentation of text in images.[0003] 2. Background Art[0004] Existing work on text recognition has focused primarily on optical recognition of characters (called optical character recognition (OCR)) in printed and handwritten documents in answer to the great demand and market for document readers for office automation systems. These systems have attained a high degree of maturity. Further text recognition work can be found in industrial applications, most of which focus on a very narrow application field. An example is the automatic recognition of car license plates.[0005] Proposals have been made regarding text detection in and text extraction from complex images and video. However, as can be seen from reading their descriptions, they are each non-general in some aspect. Further, some do not involve removal of the localized text from its background.[0006] Accordingly, a ...

AI Technical Summary

Problems solved by technology

However, as can be seen from reading their descriptions, they are each non-general in some aspect.

Segmenting and recognizing text in the non-text parts of web pages is also an important issue.

Existing text segmentation and text recognition algorithms cannot extract the text.

Thus, all existing search engines cannot index the content of image-rich web pages properly.

An input layer of 30.times.15 neurons achieved not better classification results, but was computational more expensive.

On the other side, using an input layer with less than 10 rows resulted in substantially worse results.

Again, using more hidden neurons did not result in any performance improvements, while using only one increased the false alarm rate by a factor of three.

While it is straightforward how to get examples for different types of text, it may be more difficult to get a representative non-text set.

The composition of the training set may seriously affect a network's performance.

Images