Method, memory and device for identifying and positioning underline in text image

Abstract

The invention discloses a method, a memory and a device for identifying and positioning an underline in a text image, and relates to the technical field of OCR (Optical Character Recognition). The method is characterized by comprising the following steps: identifying characters in a data picture, and obtaining the pixel width of each identified character; classifying characters with adjacent pixel widths into one class, and calculating a mathematical expected value of each class to obtain a character pixel width array; establishing a beveled accumulator array based on a variable intercept b value and a slope k value; traversing each pixel of the data picture to obtain a beveled accumulator array after pixel coordinates are accumulated; generating unit line segment pixels based on the character pixel width array, and combining the pixel quantity with continuous pixel coordinates in each component of the oblique accumulator array equal to the pixel coordinates of the unit line segment pixels into a line segment representation equation; and combining the line segment representation equations in the beveled accumulator array component to obtain the line segment representation equations of all underlines, and the method has the advantages of high accuracy and suitability for small scenes.

Description

technical field [0001] The invention relates to the technical field of OCR identification, and more particularly, to a method for identifying and locating underlines in text images. Background technique [0002] In the process of image text processing, it is often necessary to identify the underline of the data, and then fill in the required content on the underline. [0003] In some existing technologies, straight lines in the data are often detected by Hough transform, but some characters have horizontal lines similar to underscores, which have a relatively large impact on the accuracy. see attached figure 1 , for example, because the word "check" has a horizontal line at the bottom, it will cause errors in the test results. [0004] In addition, if figure 2 As shown, there is a method of extracting underline through a complex convolutional network in the existing technology. Although the method of extracting through a convolutional network has a high accuracy, the con...

AI Technical Summary

Problems solved by technology

[0003] In some existing technologies, the straight lines in the data are often detected by Hough transform, but some texts will have horizontal lines similar to underlines, which will have a relatively large impact on the accuracy

See attached figure 1 , for example, because the word "check" has a horizontal line at the bottom, it will cause errors in the detection results

[0004] In addition if figure 2 As shown, in the existing technology, there is a method of extracting the underline through a complex convolutional network. Although the accuracy of the extraction method through the convolutional network is high, the construction of the convolutional network is complicated and is not suitable for some small scenarios.

Images