Tips and Tricks for Using In-Sight ViDi Read
In-Sight® ViDi™ Read is an optical character recognition (OCR) tool that leverages deep learning to decipher deformed, skewed, and poorly etched codes for a range of applications. Combined with the In-Sight D900 camera, this powerful OCR tool significantly reduces development time, even under challenging conditions. Application setup is easy using In-Sight ViDi Read. All that you need to do is:
- Define the region of interest.
- Set character size.
To fully maximize the tool, however, you must go beyond the default configurations. Learn how to lower cycle time and increase throughput using these tips and tricks for the ViDi Read tool.
Tip #1: Check your model.
The two most common models when using the ViDi Read tool are the Node model and the RegEx model. These two are not always interchangeable and may perform differently than one another depending on the application.
A Node model is the default option when a new model is created. This can be a robust option for OCR as it can handle a variety of patterns, such as curved paths and multiple lines of text. Robustness, however, can come at the cost of processing time. If your application has linear text on a single row, using a RegEx model can yield a speedup of 70% over an equivalent Node model.
Tip #2: Don’t be afraid to change sampling density.
The Sampling Density is a runtime parameter that determines how many samples of your image In-Sight ViDi Read will process in order to return a result. While the tool defaults to a sampling density of four, changing this number may not impact the accuracy of your application. Lowering the sampling density by one can yield a speedup of 80% over an equivalent RegEx model with a sampling density of four while impacting your score by less than 3%.
Tip #3: Dial in your feature size range.
In In-Sight ViDi Read, the term “Size Range” is used to describe how big or small you expect your characters to be. A range is automatically set to make sure all expected characters are included. Keeping this default Size Range is useful for when you have variances in your font and character dimensions. If your character size is uniform though, changing the size range to resemble your expected threshold more closely can yield a speedup of over 35%.
Tip #4: Mix and match your optimizations.
When these tips and tricks are used together, a RegEx model with decreased feature size range and sampling density can be almost three times faster than a standard Node model. Still, it is important to pick the optimizations that are right for your application.
Not all these techniques for improving performance will be applicable to all situations, yet each individual improvement can yield tremendous benefits to the runtime of your application. No matter what tips you follow, always make sure to test and validate the performance of your system before deploying.