Image recognition, training and searching method, device and equipment and medium

Abstract

The embodiment of the invention provides an image recognition, training and searching method and device, equipment and a medium so as to improve the accuracy of character recognition in an image. Themethod comprises the steps that an image is acquired, and the image comprises text information; a recognizer is adopted to recognize the image, character area information in the image is obtained, andthe recognizer conducts training determination on character image data obtained in different environments on the basis of a full convolution network; and characters in the image are determined according to the character area information. Characters contained in various images can be recognized, and the recognition accuracy is improved.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to an image recognition method and device, a recognizer training method and device, a search method and device, an electronic device, a storage medium, and an operation system. Background technique [0002] The research on recognizing English symbols and numbers by computers began in the late 1950s, and since the 1970s, research on the recognition of printed and handwritten characters has begun at home and abroad. [0003] There are many methods for character recognition at present, but in natural scenes and other situations, characters usually undergo transformations, such as rotation, perspective, affine, etc. However, most of the existing text detection methods cannot realize the detection of text lines well when the text undergoes shape transformations such as rotation, perspective, and affine. Taking the text recognition method based on Maximally Stable Ext...

AI Technical Summary

Problems solved by technology

However, most existing text detection methods cannot detect text lines well when the text undergoes shape transformations such as rotation, perspective, and affine.

Taking the text recognition method based on Maximally Stable Extremal Regions (MSER) ​​as an example, if the text contrast is very low, the background area is complex, or there is a large deformation in the text area, the recognition result of the MSER method is relatively inaccurate.

Images