An Arbitrary Shape Text Detection and Correction Method
A text detection and arbitrary shape technology, applied in the field of text detection, can solve problems such as detection results are easy to stick together, candidate boxes cannot capture super long text features, etc., and achieve the effect of improving recognition ability
Active Publication Date: 2022-07-19
CHENGDU UNION BIG DATA TECH CO LTD
View PDF16 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and create a text detection and correction method of any shape. The present invention solves the weakness that the candidate frame cannot capture super-long text features through the method based on instance segmentation; The method of shrinking and labeling the text boundary area solves the problem of dense text areas and the detection results are easy to stick together; for the instance segmentation results, the text box is constructed by using the third-order Bezier curve, and then the text instance is extracted through the text box correction
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment 1
[0065] like Figure 1 to Figure 11 As shown, an arbitrary shape text detection and correction method of the present invention, such as figure 1 As shown, the method includes the following steps:
[0066] S1: Label the text image sample data set, construct a neural network text detection model based on the training label, and train the text detection model; Step S1 includes the following sub-steps, such as figure 2 and image 3 shown:
[0067] S11: Data annotation is performed on the text image sample data set, and the data annotation is as attached figure 2 and image 3 As shown, mark the text area of each text line, and make the text area label of each text line through the first border;
[0068] S12: The text core area of each text line is obtained by shrinking the text area of each text line by the contraction and expansion ratio α, and the core area label is made by the second frame, where α<1; such as the core area map;
[0069] S13: The outermost boundary i...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a text detection and correction method of any shape. Detect the core area and boundary area of each text line in the text image to be detected, and obtain the binarized and updated core area by calculating the core area and the boundary area; filter the core area with a higher score, and expand it , obtain the text area; determine the upper and lower boundary points of the obtained text area; fit the Bezier curve control points to the upper and lower boundary points of the text area respectively, and calculate the upper and lower two lines of each text line. Bezier curve; straightens the text of any shape enclosed by two Bezier curves of each text line into rectangular text. The invention realizes that the text of any shape can be detected and corrected, and the correction effect is good.
Description
technical field [0001] The invention relates to the technical field of text detection methods, in particular to a text detection and correction method of any shape. Background technique [0002] With the development of deep learning technology, the OCR recognition framework based on deep learning has been widely used in the industry. Currently, OCR text detection adopts methods based on candidate boxes and instance segmentation. The text detection method based on the candidate frame first defines several rectangular frames of different sizes for the anchor points on the feature map; then the candidate frame is determined by the frame regression of the rectangular frame and the prediction score of the rectangular frame; Maximum suppression to determine the text box. The text detection method based on instance segmentation first determines whether it contains text by judging the categories of all pixels on the feature map; then, the text region is finally determined by calcu...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G06V30/414G06V10/22G06V30/148G06V10/80G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06V30/414G06V10/22G06V30/153G06V30/10G06N3/045G06F18/253
Inventor 不公告发明人
Owner CHENGDU UNION BIG DATA TECH CO LTD