What is SmartReco-L?
- is a high-quality, flexible and fast ICR/OCR software for the recognition of machine-printed or handwritten (constrained handprinted) lines in binary images.
- can also process color and grayscale images using the special "SmartReco-L Image-File-Interface".
- calls the single character recognizer SmartReco-C internally.
- does not have to be painstakingly trained by the user, but has powerful classifiers which are developed and configured by RecoScript.
- is not an end user product, but a subprogram library for software developers.
- is based on complex mathematic transformations and provides a very high standard of recognition reliability.
- has been implemented in ANSI-C. Java and C# developers are supported by relevant wrappers.
- features three functions which can be very easily integrated, and the use of which resembles
file handling with fopen, fread and fclose:
- srl_open opens a handprint or machineprint classifier,
- srl_read reads a handwritten or machine-printed line with the aid of the classifier,
- srl_close closes the classifier and frees the corresponding memory.
- was designed platform independent and is now available as a static or dynamic library for Windows/X86 (32-bit), Windows/X64 (64-bit), Linux/X86 (32-bit) and Linux/X64 (64-bit).
The following diagram illustrates the call of the function srl_read in a typical SmartReco-L application (form reader):
The task in hand is to process a stack of similar binary images (of forms, documents etc.), where each image contains several OCR read areas (in our case a machineprint numeric and a handprint alphanumeric read area).
First all required classifiers are opened with srl_open.
All images are then read with srl_read, where the respective image, the coordinates of one read area, the corresponding classifier handle and (optional) several configuration parameters are to be passed per call. srl_read generates a result data record for each character of the line, which among others contains a list of possible unicodes with corresponding quality values (probabilities). Using its context knowledge and the result data records of the single characters, the program then forms the read result of the line (e.g. "ROUTE 65").
When the stack has been processed, the open classifiers are closed with srl_close.