Multi-category text detection system and bill form detection method based on system

Abstract

The invention provides a bill form detection method, which comprises the following steps of: inputting a pre-processed bill form picture into a multi-category text detection system, and respectively generating a central point graph, a category graph and a distance graph; searching a central point in the central point graph based on an extreme point detection method so as to determine the positionof a semantically independent field; determining the size of each semantic independent field in the distance map based on the searched central point, thereby determining a candidate box; and based onthe candidate box, determining the category of the candidate box in the category graph by adopting a voting mechanism. Compared with the prior art, the method has the following beneficial effects: based on the idea of center point detection, post-processing of non-maximum suppression (NMS) can be avoided, so that the process is simplified, the prediction process is simple, the speed is high, the detection accuracy is high, and the robustness is good.

Description

technical field [0001] The invention relates to the technical field of intelligent detection, in particular to a multi-category text detection system and a bill form detection method based on the system. Background technique [0002] Text recognition is a long-standing problem, and with the rise of deep learning, most related tasks have been well solved. However, there is a special type of data whose text is organically combined based on semantic information, called bill form-like data, which needs to extract the required information from the visual and semantic levels. [0003] Bills and form-like data play an important role in daily life. As one of the original accounting documents, it is the carrier for recording the content of economic activities and an important tool for financial management. There are many types of form-like data in different forms, which can be roughly divided into bill-type data and form-type data. For bill data, the most common ones are value-adde...

AI Technical Summary

Problems solved by technology

The disadvantage of this type of method is pixel aggregation: since pixel aggregation is performed based on the information of each pixel, the final result will be largely affected by a single pixel

First of all, the heat map needs to determine the category of each pixel according to the set threshold, so the heat map of the smallest scale is not uniform, and different thresholds will get different areas of seed points, which leads to large scales in the process of gradual expansion. Account for a small possibility, resulting in an inaccurate bounding box

Secondly, limited by the expressive ability of the small-scale heat map, when the text is close to each other, the heat map will merge, so it is easy to merge incorrectly; when the text is too long, the heat map will appear in a thin line shape, which is very prone to splitting

Images