Chinese detection method based on unsupervised learning and deep learning network and system thereof

Abstract

The invention discloses a Chinese detection method based on unsupervised learning and a deep learning network and a system thereof. In the invention, an unsupervised learning method is used to train a depth convolution nerve network, and then the network is used to classify each candidate character area in a nature image and finally text row polymerization is performed on the area which is classified as a character and a character area in the image is detected and acquired. In the invention, a super extraction capability of a deep learning network to an image characteristic is grasped and a high training capability of the unsupervised learning is used so as to realize character area positioning and segmentation aiming at a Chinese character characteristic training depth convolution nerve network. The method is simple and effective. Aiming at a character characteristic, a depth-convolution-nerve-network unsupervised learning method is established; and on an aspect of character detection, good pertinency is possessed so that high initiative and accuracy are achieved.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a text detection method and system based on an unsupervised learning deep learning network in natural scene images. Background technique [0002] Text is an important feature in many applications of computer vision. The text in the image contains a lot of useful information, which is crucial to the understanding and acquisition of visual content. The main purpose of text extraction is to convert text images into symbolic forms, which facilitates modification, retrieval, utilization and transmission. Text localization is an important step in text extraction. [0003] Text localization is the precise positioning of the text position in the image. The text localization method based on extremum connected domains first represents the image as connected domains one by one, and then proceeds from the structural analysis, marks the text lines through the merging metho...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to propose a Chinese detection method and system based on unsupervised learning and deep learning network in natural scene images, using text region feature extraction based on deep learning and classification methods can overcome the above problems and improve the recognition effect

Images