Scene text detection method based on stroke width transformation and convolution neural network

Abstract

The invention discloses a scene text detection method based on stroke width conversion and convolution neural network, which relates to the field of scene text detection. The method comprises the following steps: preparing a training data set; training a text binary classifier based on a Bootstrap strategy by the convolution neural network; training a text binary classifier based on a Bootstrap strategy by the convolution neural network. Obtaining candidate text regions from the image by using the maximum stable extremum region algorithm; classifying the candidate text region using the text binary classifier; in the candidate text region, the candidate characters are obtained based on the stroke width transformation algorithm, and the candidate characters are filtered through the geometricconstraints. The method has the advantages that the training based on the Bootstrap strategy enriches the quantity and quality of sample images; Stroke width transformation algorithm is used to improve the detection performance based on the candidate text region, and the detection level is determined to be character level. The region-based algorithm and a large number of Chinese training samplesenable the method to detect Chinese text effectively.

Description

Technical field [0001] The invention relates to the field of scene text detection, in particular to a scene text detection method based on stroke width transformation and convolutional neural network. Background technique [0002] In unmanned driving technology, it is an important task to perform three-dimensional environment modeling through perception technology. There are many related modeling data in real road scenes, such as text information in traffic signs, license plates, road signs, and billboards. The text detection and recognition in natural scene images can be used to automatically extract text information, which is one of the important research directions in computer vision. In recent years, research scholars have made certain breakthroughs in their research and built a series of evaluation databases. However, due to factors such as changeable image scenes and diverse texts, there are still many challenges in text detection and recognition in scene images. [0003] ...

AI Technical Summary

Problems solved by technology

[0006] The present invention provides a scene text detection method based on stroke width transformation and convolutional neural network, the main purpose of which is to solve the above-mentioned problems existing in the detection of scene text by existing detection methods

Images