A Method of English Text Detection with Text Orientation Correction

Abstract

The invention belongs to the technical field of image processing, and specifically relates to an English text detection method with text direction correction; the method includes: respectively performing maximum stable extremum region detection on each channel of an English text image to obtain a candidate text region; Establish a classifier based on the convolutional neural network model, filter the wrong candidate text area, and obtain the preliminary text area; use the double-layer text grouping algorithm to group the preliminary text area; correct the direction of the grouped preliminary text area, so as to obtain Correction text; The present invention adopts a kind of enhanced multi-channel MSER model: to obtain finer text area; Introduce parallel SPP-CNN classifier to better distinguish text area and non-text area, can handle the image of arbitrary size, and The pooling feature can be extracted at multiple scales, so that more features can be learned through the multi-layer spatial information of the source image; the invention can deal with slightly inclined scene text.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to an English text detection method with text direction correction. Background technique [0002] Text in natural scene images has accurate and rich information, which is of great significance for image analysis, image-based translation, image search, etc. Over the past 20 years, researchers have proposed several methods for detecting text in images of natural scenes. There are many content-based multimedia understanding applications such as automatic visual classification, image retrieval, assisted navigation, multilingual translation, object recognition, and consumer-oriented applications. [0003] The key problems faced by scene text detection are: (1) Texts in document images have regular fonts, similar colors, uniform size and uniform arrangement, but even in the same scene, texts in natural scenes may have different fonts, colors , scale and orientation. (2) The ba...

AI Technical Summary

Problems solved by technology

[0003] The key problems faced by scene text detection are: (1) Texts in document images have regular fonts, similar colors, uniform size and uniform arrangement, but even in the same scene, texts in natural scenes may have different fonts, colors , scale and orientation

(2) The background of natural scene images may be very complex

[0007] These methods distinguish text areas from non-text areas according to the rules or characteristics that distinguish text areas from non-text areas. Although these methods can detect text, they lack the correction of English text and the discrimination effect on italic text. Not good, the recognized text has severe separation due to the slant of the words

Images