TechVision's innovative products grew out of research and development in the areas of document image analysis as well as automated image data capture. Our research in these areas continues to lead to exciting product development.

Below are brief descriptions of TechVision AccuForm recognition engine, the core technology used in our solution products, and is also made available for developers and system integrators who want to add intelligent forms processing capabilities into their own applications.

• Form Recognition and Identification
• Form Registration
• Data Image Extraction
• Optical Mark Recognition (OMR)
• Barcode Recognition
• Linear Mark Recognition (LMR)

Form Recognition and Identification

In AccuForm, form recognition and identification are based on advanced neural network and fuzzy logic technologies that require neither user input nor pre-knowledge about forms. Form documents are identified via global fuzzy analysis that compares similarities of form GLS (Graphics & Line Structure) between the document forms and a target reference.

Reference (1040A) In this example, similarity values (0 - 1.0) between each incoming form and the reference is generated automatically using global fuzzy analysis. By setting a similarity threshold, say >0.85, we can easily identify all forms that are in the same type as the reference (Form 1040A). (Form 1040A) Similarity = 0.98 (Form 1040C) Similarity = 0.56 (Form 1040EZ) Similarity = 0.24

^ Top

Once the form has been identified, a complete registration process must be carried out to find the positional difference between the reference form and the input form. In AccuForm, the following registration parameters are detected automatically for each incoming form, which requires neither user input nor pre-printed anchor marks on forms:

1. Skew angle, i.e. the degree of rotation difference between the reference form and the incoming form.
   
   
2. Orientation, forms may be scanned or faxed in different orientation. AccuForm can detect and correct form orientation, e.g., whether it's in normal position or upside down.
   
   
3. Scale factor, forms may be scanned in different resolution (dpi). AccuForm can detect the size difference between the reference and the incoming form so that the possible variations in scan resolution can be adjusted automatically.
   
   
4. Offset, i.e. the translation required to align the incoming form with the reference in both horizontal (x) and vertical (y) directions.
   
   

Reference (1040A) To extract data, forms must be registered with the reference so that the positional difference between reference regions and real data on input forms can be compensated. As shown in this example, the four registration parameters for each input form can be detected automatically using AccuForm recognition engine, which doesn't require any pre-printed anchor marks or user input on forms. Skew Angle = 2.75 Orient: Normal Scale = 1.08 Offset: X = 45, Y= 7 Skew Angle = 0.43 Orient: Normal Scale = 1.27 Offset: X = 8, Y = -9 Skew Angle = -3.10 Orient: Upside Down Scale = 0.91 Offset: X = -16, Y = 38

^ Top

After form registration parameters are detected for each input form, data images can be extracted by applying the following geometric transformations on each data region defined on the reference form:

       DataImage = T_offset * T_scale * T_orient * T_deskew * ReferenceRegion

Reference Region (Before Registration) Extracted Data Image (After Registration)

The data images extracted from input forms are then sent to different recognition components for further processing, depending on the data types. Accuform has its own build-in OMR (Optical Mark Recognition) and barcode recognition engine to read extracted check boxes and barcode symbols. For other data types, such as text, numerical number, etc., it offers plug-able components (optional) at user's choice. Developers can also integrate their own recognition components to overwrite any existing recognition functions through the AccuForm API.

^ Top

This recognition module is used for recognizing optical marks in extracted check box zones. Typical application areas are in questionnaires, educational tests and in reporting or ordering sheets, where the documents to be processed are form-like and filled with check marks for selected choices.

OMR Results Marked: 1, 4, 5, 6; Unmarked: 2, 3, 7, 8

The frame of optical mark zone can be a rectangle, a hexagon, a circle, or an ellipse. It can be filled in by any method (x, tick, hatching, etc.). The frame may be visible or invisible (scanner dropout) in the image sent for recognition.

^ Top

AccuForm has a reliable built-in bar code engine that can recognize most popular bar code symbologies used in business documents, including Code 39, Code 128, Codabar, Int-2of5, UPC-A, UPC-E, EAN-8 and EAN-13.

Bar Code Symbology UPC-A Digital Code 067584930123

The recognition engine does not require predetermined bar code information, which means that not only the bar code type and orientation can be detected automatically, but the bar code skew, ink, paper and scanner variations can be compensated automatically as well.

^ Top

Linear Mark Recognition

Linear mark is another special data type suitable for automated data recognition, in which a user-drawn vertical bar crosses a horizontal line to indicate where the user feels within a given range. The value of the linear mark is defined as the length from the beginning of the horizontal line to the user-drawn mark position over the total length of the line, which is automatically captured and linearly interpreted as a floating point value between 0.0 and 1.0.

Below are examples of using linear marks in customer survey and medical treatment forms:

LMR Value = 0.8

In the above example, the user-drawn mark value is recognized as 0.8, which indicates that the customer feels 80% satisfied with the provided service.

LMR Value = 0.6

In this example, a patient feels a "0.6 headache" on a given scale of none (0.0) to strong headache (1.0).