ABOUT VEDA NeurOCR 3.0

The VEDA NeurOCR software is currently using a not-Neural but somehow alike (Artificial Neural Networks - ANN - inspired) OCR / ICR engine. It was designed and developed to become a useful tool for those who have a large amount of printed text to feed their computers with, and want to do it in a real automatic way. VEDA NeurOCR is a professional product, which is the result of hundreds and thousands of theoretical and practical research, programming and testing hours. It was developed starting from an original approach in the Neural-like Models and Classifiers field. Version 3.0 doesn't only represent a 16 to 32-bit translation of the former 2.0 one, but also comes as an almost new product, with lots of changes and improvements, both at the architecture, algorithms and (knowledge) data structures levels, and also at its user's interface "look and feel" and ergonomy ones.

• When using VEDA NeurOCR, we usually assume that you previously scanned the original document(s) you have to "feed" the computer with, and have saved it (them) as image file(s). The input image file formats accepted by VEDA NeurOCR are: black/white, pixel per bit, BMP, PCX or TIFF - either compressed or uncompressed. Any scanner accompanying acquisition software has at least one of the above mentioned image file formats among its image saving options. A scanning resolution of 300 dpi. (400 dpi. for texts written with font size less than 8 points) is recommended (100% scaled).
• A TWAIN acquisition interface to control devices (i.e. a scanner) designed and made in compliance with the TWAIN standard, is also provided by VEDA NeurOCR V.3.0., so that document images can also be directly acquired while using the program, "online". Scanned images can be stored (and will then be used) as black/white, pixel/bit, PCX files.
• The original (scanned) image can be rotated by 90, 180 and 270 degrees, and/or "negativated" (reversed b/w <-> w/b) from the program, if necessary.

VEDA NeurOCR uses its "knowledge" and other capabilities to "read" the text contained in the input BMP, PCX, or TIFF (either uncompressed, or compressed) image file(s), and will save it into correspondent ASCII text file(s).

• VEDA NeurOCR normally provides formatted ASCII text output (the recognition result follows, as much as possible, the original document layout - improved module ).
• The output text can also be obtained as "decolumnized" ASCII (columns and/or designated blocks of text are converted to successive paragraphs).
• A general DTP/WP(*) appropriate paragraph oriented text format (without <CR><LF> at the end of each line, but only at the end of paragraphs), is also available as an export option for the recognized text.
[(*) DTP = Desktop Publishing, WP = Word Processor]

• VEDA NeurOCR comes with some default (already built) font oriented knowledge bases (new format). These ones were trained on printed samples, written with the main families of fonts: Courier, Dutch (Times Roman), Swiss (Arial, Helvetica), 9 pins matrix printer and mechanical typewriter.
• It is also trainable. VEDA NeurOCR can learn and recognize any other new character and font (typeface, style, size) you want (new algorithms and data structures, improved modules). The training is interactive, fast and easy. It is always performed off-line, so, the recognition process can really be a full automatic one. Please note that, even if, in the same knowledge base, characters written with more than only one font may be learned without any problem, it is still recommended to keep each knowledge base oriented on only one font typeface and to use an appropriate name for it.
• VEDA NeurOCR is able to "read" multi-font written documents (if the knowledge bases for all these fonts are available).
• VEDA NeurOCR can even be considered as an "omnifont" -like OCR system. This means that, with a multi-font configuration setup, and with the default and the most (maximum 30 at once!) of the user trained and customized knowledge bases selected, VEDA NeurOCR can directly recognize almost any new source document, no special new settings and/or training being needed.
• It can be also considered "multi-lingual", because it is able to "read" text from documents printed in a lot of languages (mainly based on Latin-like alphabet). In fact, it can learn any graphical sign for which an ASCII correspondent exists. The learning process is natural (based on examples), fast and easy.
• VEDA NeurOCR can work in "batch processing" mode, using many different input image files in each batch session (improved module).
• VEDA NeurOCR automatically detects and skips graphics (tables, boxes, columns separators, pictures or other lines and graphical elements) in the image source and doesn't reproduce them in the recognized text file. A lines filtering option is also provided, allowing to physically erase (long) lines in an image when this can improve the segmentation and recognition (new module).
• VEDA NeurOCR V.3.0 also allows off-line interactive definition/selection of some regions/zones (of the current source image), that will be further used for recognition or/and training, if desired (improved module).
• It can obtain good recognition ratios even on poor quality input documents (obtained from 9 pins matrix printers or old mechanical typewriters).
• VEDA NeurOCR V.3.0 solves in a new more efficient and elegant approach the segmentation of physically connected ("in touch") characters, mainly encountered for Dutch- and Swiss-like fonts. It also splits, in almost all the cases, the "in touch" successive rows.
• It contains an integrated Text Editor (improved module) with which the recognized text can be analyzed, edited (corrected) and/or printed. It can also be exported from the Text Editor, in a general Word Processor paragraph oriented text format.
• VEDA NeurOCR provides a brief (baloon type) "Command Help" associated with the buttons that starts its main functions. It also has an integrated "Help Viewer" (improved module) through which the "User's Manual" text can be displayed at any time.

• The recognition ratio can reach up to 99.9 - 100 %. Its normal average value is about 98.5 - 99.5 %.
• The recognition speed can reach thousands cps. (characters/second). It is strongly dependent on:
  - the computer's CPU type and frequency,
  - the complexity of the source image (scanned document),
  - the number and the size of the used knowledge base(s).

VEDA NeurOCR can be run on any PC. However, it is recommended to use a configuration with at least Pentium II/200MHz, and 128 MB RAM on board. Minimum 10.0 MB free hard disk space must be available. VEDA NeurOCR's speed performance directly (and strongly) depends on the CPU type and frequency; e.g. on a Pentium IV/1,5GHz PC, substantial (400 - 500%) speed improvement can be observed compared with a Pentium II/200MHz.

For document image acquisition, at least a 300 - 400 dpi, line-art mode (black/white), A4, TWAIN compliant, flatbed scanner is recommended (and, usually, sufficient).

DEMO DOWNLOADING (aprox. 1.25 Mb)

A FREE evaluation (DEMO) copy of VEDA NeurOCR 3.0 (the latest version for Windows / 32-bit, also running on 64-bit!) can be downloaded from here.

[ Top of page ]