Accounting documents, perhaps, for all recognizers are on a special account. It is this class of documents that presents a special challenge to automatic data entry systems. Why? First, the documents are complex: small print, a large number of tables, inserts. Secondly, such documents are always accompanied by seals, signatures and other “joys” of office work. Thirdly, there are simply a lot of such papers in any company – there is always something to recognize. New innovative products are released with an enviable frequency that “solve” the problem of entering data from accounting documents. Highly loaded servers, distributed data recognition systems, RPA, robotic services … There are many proposals, but the task is still relevant. Why?
To be honest, only that data recognition system is needed that solves the task: a) reliably, b) efficiently, c) quickly and, of course, without “leaking data” to the side. When introducing a really useful recognition system that will solve the problem posed (unless, of course, the problem was not to “master the given budget”) there can be no compromises. All three characteristics are equally important. In support of these words, let’s just look at a few successful and unsuccessful use cases of recognition systems.
Barcode recognition works great and is used everywhere. And we are talking not only about one-dimensional barcodes with which all goods are labeled, but also about two-dimensional ones (QR, AZTEC, DataMatrix, etc.), which make it easier for us to pay taxes, follow interesting links, etc. We are so used to systems recognition of barcodes, which has already ceased to treat them not so much as recognition systems, but we interpret them as a simple and convenient way to enter data.
Face recognition, no matter how dangerous it is from the point of view potential fraud, began to be used on an industrial scale only when it really began to work quickly and reliably.
Recognition of passports and bank cards it has already become a “must have” for all banking applications and services for the sale of tickets, allowing you to order the service accurately and in a second. Even my grandmother on her simple Android stopped entering the details of her handwritten, by the way, passport.
Voice recognition is still another fun toy in smart devices (speakers, watches and other gadgets).
Recognition of traffic elements. Sometimes it seems that the key achievement in this area of recognition is to “ride on the ears” of consumers, talking about the unique unmanned vehicles that are about to appear on all the roads of our country. In fact, I get into a Volvo, BMW or Kia (underline the necessary one), with a full set of smart electronics and the first thing I do is turn off all this idle “stuffing”, where even road signs are recognized crookedly.
Recognition of accounting documents. There is on the showcase of every vendor of recognition systems. Not established from the word “absolutely” by the majority of consumers.
Of course, we are interested in the last item on this list today. What is missing for successful document recognition? Each existing accounting document recognition product lacks one of the three necessary qualities of an industrial recognition system indicated at the very beginning of the article. Let’s talk about business cases, for what purposes the accounting document recognition system is used.
Systematic entry of accounting documents in the back office and entry of recognized data into accounting systems.
Online processing and analysis of documents.
In both cases, data is entered, but there are fundamental differences. In the first case, it is possible to provide almost any level of input images (at least 600 DPI scans). Also, there are no strict requirements for the time and method of data processing. A successful solution to this problem occurs through the use of existing recognition systems and data entry services (of course, if you don’t care how much money you spend on equipment, services and salaries of employees serving such “automation”). By the way, it is the first case that provides those customer units who “successfully”, at any cost, solved the problem of automating the entry of accounting documents.
Consider now the second case. As an example, let’s imagine a logistics warehouse, where the freight forwarder driver, in accordance with the FRT or TORG-12, transfers all the goods to the storekeeper for safekeeping. What equipment, besides a ballpoint pen, is at hand at such moments? The maximum is a data collection terminal or some kind of tablet.
Another example. Field employee (auditor) who works, in fact, on the client’s premises. Agree, in the conditions of such “field” work, every minute counts (remember how much the working time of such field workers costs), which sets a high bar for the permissible recognition time. And how much does it cost to make mistakes when entering data from important accounting documents?
We are in Smart Engines did not release the system on the market for a long time recognition of accounting documents… I didn’t want to offer a semi-finished product. We collected the technological base: we were engaged in the development of the technology for searching for a document in the image, we built a unique subsystem for training convolutional neural networks, which allowed us to train, for example, bipolar neurons, and much more that was useful for the accounting document recognition system.
Smart Document Engine automatically extracts data from standard forms of documents, strict reporting forms, primary accounting, financial, tax, legal, notarial and other documents used in workflow, various tests and questionnaires, on scans and photographs.
The system allows you to automatically process single and multi-page documents with a fixed position of details, documents with a floating arrangement of blocks and details, unstructured text documents and blocks, tables, inscriptions or even separate lines and labels.
The Smart Document Engine additionally allows you to check documents for formalities: whether there is a signature, seal or logo, whether they are in the correct color, whether they are in the right place on the document, and to check that the inscriptions to be handwritten are indeed handwritten.
From an architectural point of view, the Smart Document Engine has turned out to be so versatile that we have not yet encountered a single document recognition task that would not naturally “fall” on the developed API, be it sales receipts or even information on the results of COVID-19.
That’s all. Use your health and recognize accounting documents normally.