TabletPC Development Gotchas Part3: Coerce to Factoid (XP vs. Vista)

swick — Sat, 03 Nov 2007 15:50:00 GMT

If you are writing (or maintaining) and application that uses handwriting recognition, please be aware of a subtle behavior difference between XP and Vista when coercing the recognition result to a Factoid (or an input scope or a word list). This may easily be missed in your application testing, but it will potentially result in unhandled exceptions from your application when consumers are starting to use it.

Here is the scoop: normally, when you assign strokes to a RecognizerContext for recognition, it will always be able to return some result - even for random scribbling. However, if you set the 'Coerce' flag to bias the recognizer towards a certain format (e.g. Date, URL, etc.) or word list (e.g. medical terms) then there is a (small) chance that it will not be able to compute a result. On Windows XP, the RecognizerContext returns a null RecognitionResult in this case. On Windows Vista, however, an InvalidOperationException is thrown.

So if you have written an application agains the XP behavior, please update your code to also catch InvalidOperationException if you expect your customers to use your app on Windows Vista. If you are writing a new app today on Vista, please be sure to also check the RecognitionResult for null if you plan to support your product on Windows XP Tablet PCs.

To clarify the issue in code, consider the following example: the goal here is to recognize a given strokes collection as URL. Now if the strokes are random enough, the recognizer won't be able to coerce it to a valid URL (with my handwriting, for example, writing "ooooo" does not yield a result).

private string Recognize(Strokes strokesToRecognize)

{

RecognitionStatus status;

RecognizerContext recoContext = new RecognizerContext();

recoContext.Factoid = "(!IS_URL)";

recoContext.RecognitionFlags = RecognitionModes.Coerce;

recoContext.Strokes = strokesToRecognize;

try

{

RecognitionResult result = recoContext.Recognize(out status);

if (result != null)

{

return result.TopString;

}

else

{

// Windows XP code path

return "Null means NO RESULT on XP";

}

catch (InvalidOperationException)

{

// Windows Vista code path

return "InvalidOperationException means NO RESULT on Vista";

}

So the bottom line is: if you are writing an application that uses handwriting recognition, you should harden your code to handle both a 'null' result as well as an 'InvalidOperationException' from calling RecognizerContext.Recognize().

Next post in this series: TabletPC Development Gotchas Part4: Semantics of GetSystemMetrics(SM_TABLETPC)

Previous post in this series: TabletPC Development Gotchas Part2: Deploying InkPicture applications to Windows XP desktops

East Asian Handwriting Recognition - the WPF way

swick — Fri, 02 Nov 2007 02:03:00 GMT

I have put together a sample that demonstrates the integration of East Asian boxed handwriting recognition with the inking and 3D features in WPF. The key, reusable component I have written here is the 'SingleLetterRecognizer" user control. It can, btw, also be used for western languages (by setting its 'Lcid' property accordingly), but i chose East Asian languages for this post because boxed, single character recognition is much more common in those languages.

As the user draws ink strokes into one of those 'SingleLetterRecognizer" boxes, the system's handwriting recognizer for the respective language starts analyzing the ink. When the analysis is complete and the recognition result is available, the control kicks off a 3D animation that flips the control around the x-axis to show the result as typed text. The user can double-click the control now to redo the input (if the result was not accurate).

The VS project for this sample is attached to this post. It requires Windows Vista Home Premium (or above). You can also run it on Windows XP if you have the TabletPC SDK and the Recognizer Pack installed.

Screenshot after ink input, prior to recognition complete (caveat: I don't actually write any of these languages, I just looked up some characters on the MSN homepage of the corresponding country and then wrote the simplest symbols I could find, stroke by stroke):

Screenshot after recognition complete, while animation is still going on:

Stefan Wick's Weblog - Development with Silverlight, WPF and Tablet PC : Handwriting Recognition

TabletPC Development Gotchas Part3: Coerce to Factoid (XP vs. Vista)

East Asian Handwriting Recognition - the WPF way