Method of quickly positioning Uyghur character in complex background image

Abstract

The invention discloses a method of quickly positioning a Uyghur character in a complex background image. The method comprises steps: a stroke key point is detected from an image based on a detectiontemplate; in combination of the detected stroke key point and a flooding algorithm, a character assembly in the image is extracted; based on similarities of the character assembly, assembly clusteringis carried out to realize construction of a character line; and by using a line classifier formed by an HOG+SVM system, HOG feature extraction on the character line and binary classification of a character line and a non-character line are carried out. The method can enhance the speed and the precision of positioning the Uyghur character in the complex background image.

Description

technical field [0001] The invention relates to the field of character positioning in complex background images, in particular to a fast method for positioning Uighur characters in complex background images. Background technique [0002] Uighur text location technology in complex background images has broad application prospects in the fields of image content analysis, image content retrieval and Uyghur OCR. Although a large number of scholars have devoted themselves to the research of character positioning in complex background images in recent years, their research mainly focuses on Chinese and English. The difference between Uyghur and Chinese and English makes the existing text positioning methods not robust to Uyghur, so how to improve the speed and accuracy of Uyghur text positioning in complex background images is the key to the research of this method. In addition, due to the flexibility of the text position in the image, the search process for candidates is very ti...

AI Technical Summary

Problems solved by technology

Such methods are faster than methods based on sliding window classification, but the process of component extraction and component classification is still time-consuming

[0004] A comprehensive analysis of the above two methods shows that in the existing text positioning framework of "candidate generation-candidate classification", the amount of data generated by candidates is large and the classification of candidates generates a lot of additional time overhead

In addition, there are currently no effective manual features to distinguish text and non-text candidates, and the feature framework learned by deep learning is too large and computationally expensive

Images