Multilingual natural scene text detection and recognition oriented system and method

Abstract

The invention discloses a multilingual natural scene text detection and recognition system and method. The system comprises a text detection module, a character recognition module and a language judgment module, the text detection module is responsible for the detection function of a text area, the character recognition module is responsible for the recognition function of various characters, andthe language judgment module is responsible for the judgment function of characters and languages. The method comprises the steps of text detection, character recognition and language judgment. According to the method, detection and recognition of text images of various different languages can be realized, and the used network model can be expanded to a new language only by modifying a part. According to the method, an end-to-end method is adopted, so that the data labeling process in the middle step and frequent data input and output can be reduced, and the cost is greatly saved. According tothe method, the improved FPN algorithm is adopted, good robustness is achieved, the candidate boxes are predicted through the method that the space conversion layer is combined with the LNMS, and theaccuracy of the prediction boxes can be effectively improved.

Description

technical field [0001] The invention relates to a system and method capable of detecting and recognizing text in natural scenes, in particular to a system and method for detecting and recognizing multilingual printed text in natural scenes. Background technique [0002] OCR refers to processing photos scanned by various scanning devices to obtain text information contained in the photos. Strictly speaking, OCR refers to text recognition for scanned documents, and text recognition for natural scenes is called STR (Scene Text Recognition), most of which are pictures of door signs, traffic signs, and advertisements. [0003] The difficulty of STR text recognition is much greater than that of OCR text recognition. There are three reasons: First, the diversity of text forms in natural scenes. In natural scenes, the size, font, color, brightness, etc. of characters are uncertain. Text Lines may be horizontal, vertical, slanted, or distorted, and multiple languages ​​may be mixed,...

AI Technical Summary

Problems solved by technology

[0007] 1. It is only applicable to the detection and recognition of Chinese and English, which has limitations and cannot be extended to other languages, and cannot meet the needs of multiple languages ​​​​in actual natural scenes

[0008] 2. There are many steps for image processing, and the detection speed is slow

[0009] 3. Random forest method is used for text classification, but over-fitting will occur when classification is performed in natural scenes with high noise

Images