Image text detection quality evaluation method and system

Abstract

The invention discloses an image text detection quality evaluation method and system, and the method comprises the steps of carrying out the text detection and visual drawing of an image, and obtaining a text detection line object; screening the text detection line objects of which the probabilities are smaller than a preset value to form a to-be-evaluated object set; performing quality classification evaluation on the visual image region of each text detection line object in the object set in sequence by adopting a deep learning classification network, wherein the evaluation result comprises'poor visual quality 'and'good visual quality'; and outputting the text detection line objects with poor visual quality to form a to-be-quality-inspected correction object set. According to the invention, the visual subjective visual quality of text detection is learned by adopting the deep learning classification network, and the accuracy is high; and compared with a threshold logic distinguishingmode, the invention is more consistent with the subjective visual perception of human eyes, the processing efficiency in the bill text detection and labeling process is further improved, and the enterprise manpower operation cost is reduced.

Description

technical field [0001] The invention relates to the technical field of text detection quality evaluation, in particular to an image text detection quality evaluation method and system, which are applicable to images such as bills. Background technique [0002] Text detection: Text detection is to find out where each text appears in the picture, usually there are only 2 categories "with text and without text". [0003] Quality evaluation: There are two branches of image quality evaluation methods: subjective evaluation and objective evaluation. Subjective evaluation uses people as observers to evaluate images subjectively, and strives to truly reflect human visual perception; objective evaluation methods use a certain mathematical model to reflect the subjective perception of human eyes and give results based on digital calculations. [0004] OCR (Optical Character Recognition) refers to the process in which electronic devices (such as scanners or digital cameras) check char...

AI Technical Summary

Problems solved by technology

[0009] At the same time, the quality evaluation of text detection objects can usually be achieved by setting the threshold value of the probability score, but the probability has certain limitations, and the subjective deviation from the human eye is relatively large; some images with small probability values ​​​​are also quality based on the subjective vision of the human eye. Good, even if its probability is not as good as surrounding text objects for some reason

Images