Certificate positioning method, device and equipment and storage medium

Abstract

The invention relates to an image processing technology, and discloses a certificate positioning method, device and equipment and a storage medium, and the method comprises the steps of marking the vertex coordinates of a certificate frame in a certificate sample image, and respectively generating a first thermodynamic diagram, a first scale mapping diagram and a first center offset diagram; performing prediction processing on the certificate sample image by using a pre-constructed certificate positioning model, and respectively obtaining a second thermodynamic diagram, a second scale mapping diagram and a second center offset diagram which are obtained by prediction; calculating a first loss value, a second loss value and a third loss value; calculating a loss sum according to a loss formula; performing repeated iteration training on the certificate positioning model, recalculating the loss sum, and when the loss sum is smaller than a preset threshold value, obtaining a standard certificate positioning model; and positioning the to-be-positioned certificate image by using the standard certificate positioning model to obtain a certificate frame in the to-be-positioned certificate image. According to the invention, the certificate frame can be accurately positioned from the certificate image with a complex background.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a certificate positioning method, device, equipment and storage medium. Background technique [0002] Documents are very active traffic data, and electronic documents can improve the speed of business processing. The document images uploaded in each system are of various types and the quality is uneven. Among them, there are many reflections and blurred document images caused by light spots and external dim light. If you directly evaluate the quality of the entire document image, the background outside the area of ​​the document will easily affect it and give a misjudgment result, which brings difficulties to the identification of the document. Therefore, it is necessary to locate the document before performing quality assessment on the document image. area is very necessary. [0003] At present, the methods for document location include: document location base...

AI Technical Summary

Problems solved by technology

[0003] At present, the methods for document location include: document location based on traditional machine learning method Adaboost, document location based on integral graph, cascade detector and Adaboost algorithm. The Adaboost algorithm is easily affected by complex environments, resulting in unstable positioning results and a high false detection rate

Images