A Receipt Image Classification Method Based on Color Features and Bag-of-Words Features

Abstract

The present invention relates to the technical field of images, and more specifically, to a method for classifying bill images based on color features and bag-of-words features. The present invention utilizes the classic idea of ​​Bag of Words in computer vision, that is, first extracts the SIFT feature points of each bill from the training samples and generates 128-dimensional feature descriptors, then performs K-means clustering to obtain K visual words, and Count the occurrence times of visual words for each type of bill to form a visual word histogram of this type as a feature, and finally integrate color features to form a total feature vector, which is sent to the SVM classifier for training to obtain a bill classifier model. Because the bag-of-words model does not use the color feature of the bill image, this method adds the global main color feature of the image to further improve the performance of the bill classifier. The invention only needs a very small amount of training samples and can train a bill classifier model without manually designing additional features, and the classifier has fast classification speed and high accuracy.

Description

technical field [0001] The present invention relates to the technical field of images, and more specifically, to a method for classifying bill images based on color features and bag-of-words features. Background technique [0002] In traditional bill management, manual classification of bills is often relied on. Because the number of bills to be classified is often huge, it takes a lot of manpower and material resources to complete it. Therefore, the bill automatic classification system came into being, using machine vision as the technical background to solve this problem. Class simple and repetitive classification work. Today's bill automatic classification systems need to first collect a large number of bill images as training samples, and manually design specific features for various bills, such as line segments, corners, shapes, textures, etc., and then vectorize these features into the classification Machines such as SVM for training. This classification system needs...

AI Technical Summary

Problems solved by technology

This classification system needs to collect a large number of training samples and spend a lot of effort to manually design bill features to ensure the classification performance of the trained model, so there are certain limitations.

In addition, most of the bills that the existing bill classification system can support are special financial invoices such as value-added tax and general-purpose machine-printed invoices. Versatility

[0003] Chinese patent authorization number CN106096667 authorizes a bill classification method based on SVM. This method requires manual feature design on bills in advance, such as official seal extraction and straight line extraction. This method is only applicable to a small number of bills. For most of the bills without straight lines or seals Notes cannot be classified, the method is too limited

[0004] Chinese Patent Publication No. CN107633239 discloses a bill classification and bill field extraction method based on deep learning and OCR. This method needs to first obtain seal outline features, and needs to collect a large number of seal samples as training samples for deep learning. This method is not only inapplicable Unlike most unstamped tickets, and requires a large number of training samples

Images