Character picture detection and quick matching method combining lightweight network and personalized feature extraction

Abstract

The invention discloses a character picture detection and quick matching method combining a lightweight network and personalized feature extraction, and the method comprises the steps: firstly carrying out the classification of character pictures based on a deep learning method of the lightweight network, and detecting a character picture and a non-character picture; further dividing the characterpictures into two types of character pictures, namely a complex background picture and a simple background picture; for the two types of character pictures, respectively extracting personalized feature representation picture contents; and finally, performing quick matching by using a corresponding method according to the extracted personalized features, so that the matching speed is increased while the accuracy is ensured. According to the method, the matching time can be effectively shortened, the content information of the character pictures can be comprehensively and efficiently utilized,and the character picture matching requirements with robustness and real-time performance are met.

Description

technical field [0001] The present invention takes character pictures as the research object, and proposes a fast matching method for character pictures combined with lightweight network and personalized feature extraction. First, classify Internet pictures based on the deep learning method of lightweight network, detect character pictures and non-character pictures, further divide character pictures into two types of character pictures with complex background and simple background; and then target the two types of character pictures , respectively extract personalized features to represent the content of the picture; finally, use the corresponding method to quickly match according to the extracted personalized features, and improve the matching speed while ensuring the accuracy. Background technique [0002] With the development of the Internet, smart phones, and communication technologies, the character image data on the Internet has grown rapidly. While these data bring r...

AI Technical Summary

Problems solved by technology

The character image matching technology based on deep features uses a deep neural network to detect text of any shape in the image. After detecting the character area, this method performs similarity measurement on the extracted deep features to complete the matching of character images. This method seldom considers the running time and efficiency of the algorithm, which makes it limited in the actual application environment; the character image matching technology based on local features extracts the local features of the image through the local feature extraction operator, which can effectively represent complex background character classes. Image features, local feature extraction operators mainly include SIFT (Scale Invariant Feature Transform), SURF (Speeded-Up Robust Features), ORB (Oriented FAST and Rotated BRIEF), etc., where the expression ability of ORB features is similar to SIFT, SURF, etc. Local features, but the detection speed is one to two orders of magnitude faster than SIFT and SURF, which can effectively characterize the characteristics of complex background character pictures and meet the needs of real-time matching of pictures

Considering that simple background character pictures are pure character pictures, the mainstream technology for this kind of picture recognition is Optical Character Recognition (OCR) technology. Electronic documents with consistent character content, and then use text matching technology to match character images, but OCR technology is not accurate enough for highly deformed characters, handwritten characters, etc., and requires high computing power and slow processing speed, which cannot meet the requirements Real-time matching requirements for character pictures; picture feature words are a kind of picture features that use morphological processing to detect character areas, extract relevant areas for encoding, and concatenate them in a certain order. Feature words can effectively represent the features of character pictures with simple backgrounds , and the feature extraction speed and subsequent matching speed are both fast, which can meet the real-time matching requirements for character pictures

Images