A mobile terminal system and method based on table character detection and recognition

Abstract

The invention belongs to the technical field of mobile device data processing, and discloses a mobile terminal system and method based on table character detection and recognition. According to the relationship between the geometric direction of the object in the image and the Fourier spectrum corresponding to the image, the inclination correction is carried out on the collected image to obtain the corrected image. Through OTSU Otsu method to binarize, then define horizontal bar and vertical bar to corrode the check image, dilate the operation to obtain table lines, so as to segment the imageto obtain characters; A convolutional neural network for letter and Chinese character recognition is constructed. By randomly adjusting the brightness of image, expanding the sample of contrast and font thickness, the data can be enhanced, the classification standard can be established automatically, and the adaptability of complex background can be enhanced. The invention integrates image segmentation, character detection and recognition, and realizes table character recognition scanning application program based on convolution neural network.

Description

technical field [0001] The invention belongs to the technical field of mobile device data processing, and in particular relates to a mobile terminal system and method based on table character detection and recognition. Background technique [0002] At present, the commonly used existing technologies in the industry are all focused on the research of printed character recognition, most of which are based on large-scale scanning devices or cannot be run on mobile terminals, and the character recognition algorithms used, such as nearest neighbor algorithm or template matching algorithm, exist Many deficiencies are embodied in the following two aspects: [0003] The template matching algorithm and the nearest neighbor algorithm that perform well in character recognition need to store samples or sample features, and the storage capacity increases with the number of samples. However, in order to achieve better classification results, the sample size is indispensable. , which make...

AI Technical Summary

Problems solved by technology

[0006] (1) Most of the printed characters are regular fonts. Of course, the recognition of printed characters is much simpler than that of handwritten characters, but the recognition of printed characters can no longer meet the needs of people's daily life;

[0007] (2) In the prior art, the template matching method and the nearest neighbor algorithm, which perform better in character recognition, need to store samples or sample features, and the storage capacity increases with the increase of samples. However, in order to achieve a better classification effect , the sample size is indispensable, which makes it unrealistic to realize character detection and recognition on the mobile terminal;

[0008] (3) Due to the unification of character styles and the simplification of expression forms, it is difficult to find effective characterization features, so the classification effect based on artificially designed features is not good;

[0009] (4) The current technology does not consider the integration of character recognition into the lives of the public, and it is mostly used in teaching, teaching auxiliary companies and other places, and requires large-scale scanning and testing equipment

[0011] The difficulty lies in: everyone may have a handwriting style, but different styles of fonts need to be recognized as the same character, and the types of characters are huge. There are 3755 commonly used Chinese characters alone, which means that the project is huge and complicated. However, The application on the mobile terminal means that it is necessary to use limited descriptors to describe huge and complex tasks;

Images