Difference background grid-oriented curve extraction method based on deep learning

Abstract

The invention discloses a curve extraction method for differential background grids based on deep learning. Comprising the following steps: shooting a paper report image with grid lines, positioning a curve signal area in the image, converting the curve signal area into a binary image, performing image segmentation by using a DeepLabV3 + model, and separating curve signals, background grids and other parts. Analyzing breakpoints existing in the curve signal, judging the type of the breakpoints, and repairing by adopting different modes according to different breakpoints until a complete curve signal is obtained. And performing skeletonization operation to complete curve extraction. The method is not influenced by the background grid form, complete extraction of the curve signal can be successfully realized for the background grid in any color or form, the quality of the acquired image is not required, and the curve signal can still be successfully extracted even if lens distortion or angle inclination exists.

Description

technical field [0001] The invention belongs to the technical field of computer image processing, and in particular relates to a curve extraction method based on deep learning for differential background grids. Background technique [0002] With the development of science and technology, the proportion of digital images in daily life is increasing. Compared with paper images, digital images are not only more convenient to save and disseminate, but also can be further analyzed by digital means. Especially in some medical scenarios, digitizing the paper signal maps will not only help doctors to read and judge them more accurately, but also eliminate the subjective differences caused by human intervention, and can also be used to build a rich database. Provide support for subsequent analysis systems. However, many early data still exist in the form of paper, so it is very necessary to study the digital method of paper signal map. [0003] The existing methods for extracting ...

AI Technical Summary

Problems solved by technology

Among them, the color filter method mainly uses the color difference between the signal line and the background grid, so the background grid with similar color characteristics to the signal line cannot be removed.

The threshold method uses the grayscale characteristics of the signal line and the background grid line, and cannot remove the background grid similar to the grayscale feature of the signal line.

Morphological methods, Hough transform methods, and line projection methods use the shape characteristics of signal lines and background grids. These methods have relatively high requirements for the quality of the collected pictures, and it is easy to remove the background grid uncleanly due to the problem of lens distortion.

The method of two-connected domain and weight sum cannot remove the background grid similar to the signal line according to the threshold characteristics of the signal line and the background grid

In addition, the above methods are all aimed at the digitization of color signal maps and binary signal maps, and cannot handle signal maps with shadows, and most of them only target one kind of characteristic background grid. In the actual operation process, it is necessary to According to the characteristics of the grid, different signal line extraction methods are selected, that is, there is no general curve extraction method for differentiated grid backgrounds.

Images