Multi-directional text detection method based on corner points
A text detection, multi-directional technology, applied in the field of text detection technology, can solve the problems of low positioning accuracy and high difficulty of text positioning tasks, and achieve the effect of improving detection performance and detection speed
Active Publication Date: 2021-11-09
UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF5 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0003] However, for text in natural scenes, since there are no certain prior conditions and given environmental constraints, the task of text positioning is much more difficult than that of document images.
Traditional text detection in natural scenes mostly uses sliding windows to locate text, which leads to a rather low positioning accuracy. Therefore, it is necessary to propose a text detection that does not require any prior knowledge to locate text information in natural scenes. technology
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
[0049] See figure 2 First enter the image to be detected, such as figure 2 The left of the middle;
[0050] The probability is then selected by the probability matrix, which is greater than the pixel point of the preset threshold T as the candidate angle, and determines the diagonal of the rule of candidate angle point, such as figure 2 The left is shown in left;
[0051] Next, based on steps S3, S4, a series of candidate recommendations are generated, such as figure 2 The right shown in the right;
[0052] Finally, based on the non-empty probability of the candidate recommendation, the NMS method removes the redundant candidate suggestion box, and the final detected text area, such as figure 2 The second of the second is shown.
[0053] The end-to-end training proposed by the present invention, and the training-out detection model can detect the multi-directional text in the natural scene, and in 3 public data sets (ICDAR-2013, ICDAT- 2015, coco-text, under test, F-Measure (weigh...
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 corner-based multi-directional text detection method, which includes two processing parts: firstly, region proposal regions are generated by detecting corner points, and these regions are used to estimate the possible positions of text regions, because unlike other The proposed region generated by the algorithm is a horizontal rectangle, and the present invention can generate suggested regions of various angles by using corner points, so the detection method of the present invention can detect texts at various angles such as inclination, and finally use the subnetwork proposed by the present invention to further analyze these It is recommended to train regions for classification and regression, and finally locate the text regions accurately. The present invention is used for text detection in natural scenes, and its detection performance and detection speed are significantly improved compared with the existing processing methods.
Description
Technical field [0001] The present invention belongs to the field of image processing and pattern recognition, and is specifically related to text detection techniques for natural scenery. Background technique [0002] With the popularity of various handheld devices with cameras, text detection under natural scenes has new development. More and more people can use a variety of electronic devices anytime, anywhere to shoot the things and scenes you see, record the information, such as road signs, billboards, store trademarks, etc. In the past, many text extraction methods are performed for certain specific environments and give certain prior knowledge, such as subtitle extraction in video images, often have fixed alignment directions, arranged in a row or column arrangement, text size It is fixed or varied, and the position of the appearance is also usually the bottom or on both sides of the video image, and the color of the same subtitle area is often consistent; and the label de...
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): G06K9/00G06K9/46
CPCG06V30/413G06V10/44
Inventor 马争卢欣辰解梅陶帅
Owner UNIV OF ELECTRONICS SCI & TECH OF CHINA