A denoising method and a denoising device for linear interference

Abstract

The invention discloses a denoising method and a denoising device aiming at linear interference, which relates to the computer technical field. An embodiment of the method includes: preprocessing an original image to obtain a binary image; determining a coordinate interval corresponding to each character in the binarized image; determining the coordinates of the first end point and the end point corresponding to the character according to the coordinate interval corresponding to the character; solving a noise curve expression according to the coordinates of a head-end point and an end point corresponding to each character; noise being predicted and removed using a noise curve expression. The embodiment can precisely solve the segmented expression of the noise curve, better fit the noise curve, realize more accurate character recognition, and further improve the recognition success rate of the whole verification code.

Description

technical field [0001] The invention relates to the technical field of computers, in particular to a denoising method and a denoising device for linear interference. Background technique [0002] With the rise of crawler technology, more and more businesses are beginning to rely on the data crawling capabilities of web crawlers. For security and traffic control considerations, various websites have also enabled verification code verification to block web crawlers. With the rapid development of image recognition technology, the patterns of verification codes are also gradually expanded and refurbished, from the earliest characters are correct, the size is consistent, the background is clean, and the background is discolored in the later stage. Various noises are added, and the character rotation is blurred and deformed. It can be said that , image recognition is the key to open this door, and image denoising processing is a key step in the pre-order operation of image recogni...

AI Technical Summary

Problems solved by technology

[0007] 1. For solution 1, because the verification code numbers and letters are changeable, if only simple denoising cutting is performed, the main bar-shaped interference noise will not be removed, and the data complexity is high

In order to ensure the recognition accuracy of the training model, this requires a large amount of marking data for each category, the cost of manual marking will increase, and the training time of the model will also become longer. The achieved stable recognition rate is not optimistic;

[0008] 2. For scheme 2, by Figure 4 The schematic diagram of the effect shown shows that the simple longest connected domain removal has false elimination. Not only does it not remove the linear noise, but it removes the recognizable character part. Obviously, this will eventually affect the character recognition rate.

If the noise completely overlaps with a part of the character, then the overlapping part of the character will also be removed while removing the noise, resulting in an incomplete character

If there is an intersection between the character and the noise, this part is very likely to be removed because the arc of the character curve is larger and longer than the noise, and it will also cause incomplete or even deformation of the character

Images