Prefix number identification method and system

Abstract

The invention provides a prefix number identification method and system. The method comprises: extracting the gray level image of the area where a prefix number is located, and performing edge detection on the gray level image; rotating the image; performing fine mapping on a single number in the image, tightening characteristics included in the image of a positioned number, and performing normalization processing on each number image, normalization including dimension normalization and light and shade normalization; and employing a neural network to identify a normalized number image to obtain the prefix number. The method can guarantee an operation speed, meanwhile improve identification robustness, and guarantee accuracy and practicality in practical application.

Description

technical field [0001] The invention relates to a method and system for identifying serial numbers, which are suitable for identifying serial numbers on special printed products. Background technique [0002] At present, whether it is in the fields of cash deposit and withdrawal, check payment transfer, license plate number recognition, etc., there are important requirements for the recognition of serial numbers in the area to be recognized, especially in the recognition of banknotes, which require high accuracy. , the recognition efficiency and recognition speed are also high. [0003] In the prior art, with the development of DSP technology, it is more common to realize the recognition of the serial number through the DSP platform in cooperation with computer vision technology and image processing technology. In the specific recognition algorithm, commonly used methods include template matching, BP neural network, support vector machine, etc., and multiple neural network ...

AI Technical Summary

Problems solved by technology

In the DSP identification method, it is often limited to the network transmission efficiency and the influence on the position and orientation of the banknote in the DSP identification, and the identification efficiency and the robustness of the identification algorithm are relatively poor. For example, in the patent application number CN201510702688.2 In the application, the edge is fitted through the gray threshold and direction search, and then the edge line is screened through the threshold to obtain the slope of the area. Combined with neural network training to identify the face, it is identified through progressive scanning and subsequent neural network. Serial number

[0004] Another example is in a prior art, such as the paper "Research and Implementation of RMB Sorting Method Based on Image Analysis", which has adopted the method of convolutional neural network to identify the serial number. However, in the above-mentioned scheme, only the most Simple binarization divides the characters, which cannot effectively tighten the characters, and this will directly affect the amount of data that needs to be processed in the follow-up, and directly affect the practical value of the algorithm; The size processing does not effectively tighten the preprocessed and segmented images and effectively normalize the image data, and this simple size processing will bring heavy data processing to the subsequent neural network recognition, which is extremely difficult. The subsequent identification efficiency is greatly reduced; and, the above-mentioned technical solutions also do not deal well with the impact of banknote incompleteness on banknote identification and image processing

Although the above-mentioned technical solution can theoretically achieve a certain recognition accuracy, due to its low efficiency of calculation and recognition, it cannot be well transformed into a commercial practical method, and cannot meet the speed requirements in real banknote recognition.

[0005] It can be seen that the existing technology has the following problems: the orientation of banknotes and the effective positioning of characters cannot be efficiently solved, the range of characters after recognition is relatively large, and it is easy to cause wrong division of characters, and the amount of data required for later image processing and recognition is large , which reduces the recognition efficiency; it cannot adapt well to the rapid tilt change of the banknote image caused by banknotes, and cannot correct and identify the tilt of the banknote in time; the robustness of the identification of damaged banknotes is low, and no corresponding banknote damage is provided. Identify processing method

Images