wedge
During the study and work process of researchers, they will encounter the problem of chart analysis and classification, so they cannot avoid the key technology of OCR. When you need to organize research reports or compile papers, you first need to refer to mainstream research reports, papers or think tank reports in the market. Among the reference sources, the vast majority of data materials are in image or PDF format. Manual reuse and transcription are slow and error-prone. Therefore, everyone will have a strong demand for OCR technology.
OCR technology development history
OCR is the abbreviation of Optical Character Recognition (Optical Character Recognition) technology, which refers to the use of machines to convert handwritten or printed text in images into a format that can be directly processed by computers. As an important branch of the computer vision field, a typical application of OCR is to achieve information entry through image and text recognition.
In the past 40 years of technological development, OCR has always had a strong industrial application background and is one of the few fields in the computer field that has been driven by both industry and academia from the beginning. In recent years, OCR technology has been maturely implemented in the industry, but the research interest in this field in academia is weaker than in other directions. Some people even think that OCR technology is fully mature and there is no need for more research.
However, with the gradual implementation of intelligent text processing IDP (Intelligent Document-nt Processing) in the industry in recent years, there are more and more application scenarios combining OCR and IDP. Using the perspective of semantic understanding NLP to further extend the application of OCR, many scenarios with more industrial application value have emerged.
With the proliferation of human industry and the continuous development of information technology, deep learning theory has made breakthrough progress, stimulating commercial and civilian demand to explode rapidly with the popularization of mobile Internet. Driven by the two factors of "smartphone + deep learning", the research and development of OCR technology has ushered in three new hot directions in recent years, namely:
Ⅰ-OCR is combined with intelligent text processing (IDP) to perform semantic understanding and structural analysis of unfixed format documents. It not only identifies the text itself, but also understands the layout, structure, table elements, paragraph content, etc. of the text, thereby completing the restoration and structured extraction of text element information, and is used in scenarios such as intelligent document review processing.
Ⅱ- OCR is combined with symbol recognition in professional fields, such as mathematical formula symbols, physical formulas, chemical molecular structure diagrams, architectural drawings, etc., to realize applications in professional fields, such as picture search, drawing review and other scenarios.
Ⅲ- OCR is combined with open scene text recognition (often called STR, Scene Text Recognition), such as street signs, store signs, trademark text, outdoor advertising recognition, etc., and is used in transportation, outdoor consumption, autonomous driving and other scenarios.
A brief analysis of user pain points
At present, mainstream OCR software has two methods: web page and client, which are mainly charged by the time. The main problems that affect the user experience are:
1. OCR recognition cannot cover the complete image. When OCR parses a chart, the center area of the chart occupies the main space, and the parsing accuracy is high, but the title, subscript and other content are often garbled, which hinders researchers from writing research reports and reusing charts;
2, OCR recognition results need to be stored locally and require manual classification. When more than a hundred charts are used in an in-depth research report or paper writing process, it is time-consuming and inconvenient to manually sort them one by one, and products are urgently needed to solve them.
The first type of technical application is to solve the above problems. OCR+IDP is mainly oriented to business analysis and research work. It can solve the problems encountered by most investment research workers such as insufficient analysis accuracy and unclassifiable analysis content. It is also one of WarrenQ's main product application directions. Please take a look at WarrenQ's current solutions below:
WarrenQ overall analysis solution
WarrenQ's overall analysis function is divided into three aspects: direct analysis of the research report library, user-on-demand chart analysis, and automatic analysis of cloud disk uploads.
01 Comprehensive and convenient research report library analysis ✦
WarrenQ, as a lightweight intelligent investment research platform, directly integrates the research report library function internally. When all research reports are stored in the database, all charts will be extracted and displayed separately, and NLP will be used to mark the research reports and charts, and content extraction and classification will be implemented on the production side.
When viewing the research report, users can directly view the label information of the research report and view all the extracted chart materials at the same time, which is convenient and fast.
02 Accurate and efficient OCR✦
Currently WarrenQ's OCR function supports the analysis of various charts. Pure tables, line charts, bar charts, area charts can all accurately identify data. The OCR recognition accuracy rate is over 90%, which can meet the needs of commercial offices.
The latest version of the OCR function already supports unlabeled chart recognition. Even if the data is not clearly marked in the picture, numerical estimation can be made through the Y-axis tick mark to help researchers organize data in more scenarios.
03 Cloud disk analysis, intelligent classification ✦
The production investment research content uploaded by the WarrenQ platform through the cloud disk can be intelligently labeled with various tags through AI recognition. Users can quickly identify the attributes of the investment research content based on the tags, such as industry, related products, etc., and move them to corresponding projects accordingly.
All investment research content uploaded to the platform can be viewed online, without the need for cross-platform and cross-software, and OCR recognition charts are supported.
Conclusion:
products have provided services to many financial institutions; in the future, the products will continue to work with institutions to carry out in-depth cooperation, integrate innovation, improve experience, and help researchers improve usage efficiency. The
product is currently officially online and is open to registration. Research workers and institutional users are welcome to register and use it through the URL below. Thank you!
https://www.warrenq.cn/
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-