IText - OCR screenshot recognition | Toolinbox

IText is an OCR tool that recognizes text from images.

Typical usage scenarios:

1. Extract text from scanned PDF

2. Identify text from pictures sent by friends

3. Read from any image

Product Features

1. Convenient selection of images

IText supports multiple ways to select images, making the operation very convenient.

1.1 Screenshot for Literacy

IText comes with a screenshot function, simply press the ⇧⌘ 1 shortcut key and capture any position on the screen to extract the text from the image.

Tip: The recognized text has been copied to the system clipboard, just paste it directly.

1.2 Drag and drop images to the menu bar icon to read characters

For example, when you use WeChat or QQ on your computer, if a friend sends a picture (such as a delivery note taken on your phone) and you want to recognize the text (such as the delivery note number), simply drag and drop the picture to the iText menu bar icon.

1.3 Selecting Image Files for Word Recognition

Of course, you can also select image files in the dialog box. However, I personally recommend that you directly drag and drop the image file to the iText menu bar icon for intuitive and efficient operation.

1.4 Support continuous recognition

For example, by continuously capturing images from different positions in a PDF, iText will recognize the text in sequence and automatically concatenate the results together.

2. Accurate text recognition

Have you ever had the experience of trying to extract text from an image, only to find some errors in the recognized text. As a result, the time it takes to manually correct these errors is longer than the time it takes to directly write words on the computer. Isn't it awkward?

It can be seen that the accuracy of recognition is very important, and it is also a point that I am very concerned about.

2.1 Tencent, Baidu, and Google three engines

Firstly, I excluded offline recognition libraries because their recognition capabilities are doomed to be dead and will not self enhance. Next, in the online OCR service, I compared the products of four major companies: Baidu, Tencent, Microsoft, and Google. Finally, I chose a combination of using Tencent and Baidu domestically and using Google abroad.

If you have Google faith or need to recognize other languages such as Japanese, you can right-click on the iText menu bar icon and manually select to use Tencent, Baidu, or Google OCR services. However, based on actual testing and current optimization, it is still recommended to use Tencent for Chinese; Everyone can test it themselves.

So, is the recognition accurate? Let me put it this way:

For general natural language, such as a passage in a book or a news release, the recognition effect is astonishingly accurate, even reaching 100%

For text with complex layout and special characters, such as program code and multiple-choice questions, the recognition effect is not ideal, and manual correction of the recognized results is required

For example, simply giving a vertical line, the machine cannot distinguish whether it is lowercase 'l' or uppercase 'I' (by the way, can you see the difference between the two?); In contrast, machines need to make judgments and optimizations based on context. However, for non natural language programs like program code, it is currently difficult for machines to perform semantic recognition

Is it accurate or not? Welcome to give it a try.

2.2 Innovative algorithm to further optimize recognition results

The best OCR service is character recognition; But further recognition, such as paragraph recognition, full and half corner symbols, etc., is not very good at present, which requires optimization in the later stage of the program.

At present, iText has mainly completed the optimization of these parts:

Automatically recognize paragraphs

Use full width punctuation in Chinese environment

Add spaces between Chinese and English letters and numbers

Remove excess spaces between Chinese characters and between English characters and punctuation marks

Initial capital letters in English

If you find any areas where the optimization is not good, please feel free to send me the graph and I will optimize the algorithm accordingly. Thank you in advance.

2.3 Preview the original image for easy proofreading

In the current situation where OCR technology cannot always achieve 100%, it is still necessary to preview the original image and make corrections. In iText, you can do this:

Drag the recognized window near the image

Call iText's dual column mode for recognition results: display images on the left and recognized text on the right

You can use the shortcut key ⇧⌘ R to quickly bring up the recognition result window

In this way, it is easy to manually optimize and recognize the text by comparing it with the original image.

2.4 Automatically hide recognition results

Since iText's recognition results are already very accurate and have been copied to the clipboard, there is no need to edit or copy the text after recognition. At this point, you can turn on the "Auto Hide" option as shown in the above picture, and the recognition result window will automatically hide after 3 seconds, which is very convenient.

And if there is a temporary need to edit a recognition result, simply move the mouse to the result window, and the auto hide function will be temporarily ignored. In addition, when the "always at the front" option is enabled, the window will not automatically hide.

3. Automatic translation after recognition

For example, for English PDFs, untranslated subtitles, and games that have not been translated into Chinese, simply use iText to take screenshots and automatically translate them into Chinese.

Integrated with Google and Baidu dual engines (Google is preferred, and Baidu is used when Google cannot be accessed), supporting over 100 languages.

4. Download and Payment

Free identification up to 20 times per month to fully experience product features; Subscribe to iText Advanced Edition for unlimited recognition.