|
2011-04-13
, 20:25
|
Posts: 183 |
Thanked: 33 times |
Joined on Aug 2010
|
#2
|
|
2011-04-13
, 22:45
|
|
Posts: 2,361 |
Thanked: 3,746 times |
Joined on Dec 2007
@ Berlin - Love this city!!
|
#3
|
|
2011-04-14
, 02:48
|
Posts: 356 |
Thanked: 172 times |
Joined on Jan 2010
@ Canada
|
#4
|
|
2011-04-14
, 03:42
|
Posts: 195 |
Thanked: 102 times |
Joined on Feb 2010
@ Vancouver, Canada
|
#5
|
|
2011-04-14
, 04:39
|
|
Posts: 28 |
Thanked: 3 times |
Joined on Sep 2010
|
#6
|
|
2011-04-14
, 04:52
|
Posts: 12 |
Thanked: 148 times |
Joined on Apr 2011
|
#7
|
|
2011-04-14
, 10:13
|
Posts: 141 |
Thanked: 51 times |
Joined on May 2010
@ Netherlands
|
#8
|
|
2011-04-14
, 10:34
|
|
Posts: 855 |
Thanked: 612 times |
Joined on Oct 2010
@ Germany
|
#9
|
The Following User Says Thank You to Alfred For This Useful Post: | ||
|
2011-04-14
, 10:55
|
|
Posts: 565 |
Thanked: 618 times |
Joined on Jun 2010
@ Finland
|
#10
|
MIR Translator is an application for translating text in photos taken by the N900 camera. It experiments with novel interactive techniques to perform text recognition. Currently the app supports translation to and from over 30 different languages. Video Demo: http://www.youtube.com/watch?v=VT-XsUPXCFA
Requires internet connection to work.
Package is available in extras-devel. Be sure to restart after installation if FCam drivers were not previously installed.
Maemo packages URL: http://maemo.org/packages/view/mir-translator/
The app relies on the text recognition happening off the device and in the cloud. This allows the application to support the maximum number of languages possible while keeping the install size as small as possible (~1.5 MB). It also improves response times, given a fast connection and server side.
Features:
-A viewfinder is used to capture images with the camera. The user selects text to translate.
-If the recognition is erroneous, the user is able to correct the recognition via manual segmentation, manual thresholding, and manual candidate selection.
-Supports more than 30 input and output languages.
Limitations:
-Somewhat limited performance. This is because the recognition service resides in the cloud as a web service. I am hosting the service myself, so server capacity / speed is as a result limited.
-Recognition accuracy is currently quite dependent on the quality of the input image. That is, images should be well and evenly lit and have good contrast. It's best if the text is sharp, with typographic-like fonts and non-complex backgrounds.
This project fell out of a project I was doing in school (its actually my first maemo app!). I've recently adapted it to be deployed publicly, so its in an early development phase. Naturally there will probably be some bugs, and there is much room for optimization. I don't expect the application to be powerful enough for the general case of translating arbitrary text-- but simpler cases should be do-able. Deploying it publicly is fun and I'm hoping it speeds feedback + development.
I'm keeping the web service up and running on my own student funds -- its not cheap! Any and all small donations are greatly appreciated (Donate Link: Donate). Hope you enjoy the app!
Last edited by mhsueh; 2011-04-15 at 00:31.