Document video feature extraction method, device, computer equipment and storage medium

Abstract

The invention discloses a certificate video feature extraction method, device, computer equipment and storage medium, belonging to the technical field of video processing. The method includes: acquiring an image sequence from a video containing a certificate, the image sequence including at least one frame of a certificate image; extracting multiple feature subgraphs of each frame of the certificate image according to multiple anti-counterfeiting features of the certificate; Multiple feature subgraphs are respectively matched with preset multiple image templates to obtain matching score vectors corresponding to multiple image templates for each frame of document image; form a match according to the matching score vector corresponding to each frame of document image Score matrix and mapped to a video feature vector. The invention extracts the video features of the certificate based on template matching to realize authenticity identification of the certificate, has good versatility, can take into account various features including dynamic features, and has good scalability.

Description

technical field [0001] The present invention relates to the technical field of video processing, in particular to a method, device, computer equipment and storage medium for extracting certificate video features. Background technique [0002] Video classification is widely used in many fields such as anti-fraud and terrorism detection. One of the typical applications is the authenticity identification of documents in the financial field. [0003] General video classification methods, for example, the application number is CN201711420935.5, and the patent application titled "video classification model training method, device, storage medium and electronic equipment", the application number is CN201711172591.0, and the invention name is "video classification The patent applications for "method, video classification device and electronic equipment" all require the use of deep neural networks to extract features, which further requires a large amount of training data or at leas...

AI Technical Summary

Problems solved by technology

[0004] However, in the issue of identifying authenticity of documents, due to factors such as user privacy and difficulty in data collection and labeling, usually there is not a large number of standard data sets, and the anti-counterfeiting features of different documents are also quite different, so it can be used to train algorithms. There are few videos, and it is difficult to find a suitable pre-trained model to extract features, which is not suitable for directly using neural networks to extract features

In addition, since the quality of video collected by users is usually difficult to meet a very high standard, and the accuracy of a single identification feature of a document is often low, it is necessary to use multiple features at the same time to achieve counterfeiting

However, the anti-counterfeiting features of documents are not the same. For example, some dynamic features will change with changes in light angles. If the methods used to extract various anti-counterfeiting features are very different, the system will be too complicated and not easy to maintain. and extension

Images