General questions

Recognition Accuracy

We are using your CDK package,
We observed that the recognition accuracy rate for English is less than 90%, (Separate Capital letters)
where as for other language like Spanish we get more than 95%
The input is from Pegasus Digital Pen and the Sample rate(XY coordinates) is around 40-60 Per second
Looking forward your input and support to increase the Accuracy rate


Dear Nelson,

thank you for contacting us.

We are actually quite puzzled by this behavior, as accuracy is likely to be the same from one latin language to another.

Is it possible to learn more about your configuration:
-Which resources are you using?
-Are you in isolated or cursive mode?
-Can you please provide us with an ink sample?
-Ideally, would it be possible to get a request that you post, so that we can reproduce on our side?

Thank you,

Best regards,


Dear Olivier,

For individual character we use textInputMode as ISOLATED and textInputType as CHAR but we are getting | for I for the following.
Please find the below POST,

applicationKey=XXX&textInput={textParameter: { language: en_US,textInputMode: ISOLATED,resultDetail: CHARACTER,contentTypes:[text]}, inputUnits: [{textInputType: CHAR,components: [{type: stroke, x: [10760,10758,10758,10750,10756,10760,10756,10754],y: [ 13011,12998,13009,13032,13083,13142,13195,13250]}]}]}

Please let us know, if the above POST is in correct format.
We mainly use Numeric, Alphabets,Only Alphanumeric, Alphanumeric with Symbols, Email, Single Character either Digit or Alphabet but not Symbols.


Dear Nelson,

Thanks for those details, in such case, '|' 'I' can be recognized and actually the latin capital letter I appears as second candidate:
result: {
textSegmentResult: {
selectedCandidateIdx: 0,
candidates: [
label: |,
normalizedScore: 0.67284244,
resemblanceScore: 0.7956908,
children: [
inkRanges: 0-0-0:0-0-7,
selectedCandidateIdx: 0}]},
label: I,
normalizedScore: 0.32715756,
resemblanceScore: 0.75058687,
children: [
inkRanges: 0-0-0:0-0-7,
selectedCandidateIdx: 1}]}
It could be of help to use custom resources to improve recognition accuracy according to your use case as explained in this post:
Please let us know if it helps.
Best regards

Login or Signup to post a comment